This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[PATCH] BZ #14649: Add multiarch FMA support to x86-64 libm
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: GNU C Library <libc-alpha at sourceware dot org>
- Date: Mon, 1 Oct 2012 08:14:58 -0700
- Subject: [PATCH] BZ #14649: Add multiarch FMA support to x86-64 libm
Hi,
This patch adds multiarch FMA support to x86-64 libm. Tested on
FMA machine. OK for master?
Thanks.
H.J.
---
sysdeps/x86_64/fpu/multiarch/Makefile | 31 ++++++++++++++++
sysdeps/x86_64/fpu/multiarch/doasin-fma.c | 4 +++
sysdeps/x86_64/fpu/multiarch/dosincos-fma.c | 6 ++++
sysdeps/x86_64/fpu/multiarch/e_asin-fma.c | 11 ++++++
sysdeps/x86_64/fpu/multiarch/e_asin.c | 17 +++++++--
sysdeps/x86_64/fpu/multiarch/e_atan2-fma.c | 10 ++++++
sysdeps/x86_64/fpu/multiarch/e_atan2.c | 6 ++--
sysdeps/x86_64/fpu/multiarch/e_exp-fma.c | 6 ++++
sysdeps/x86_64/fpu/multiarch/e_exp.c | 6 ++--
sysdeps/x86_64/fpu/multiarch/e_log-fma.c | 8 +++++
sysdeps/x86_64/fpu/multiarch/e_log.c | 6 ++--
sysdeps/x86_64/fpu/multiarch/e_pow-fma.c | 6 ++++
sysdeps/x86_64/fpu/multiarch/e_pow.c | 14 ++++++--
sysdeps/x86_64/fpu/multiarch/halfulp-fma.c | 4 +++
sysdeps/x86_64/fpu/multiarch/mpa-fma.c | 12 +++++++
sysdeps/x86_64/fpu/multiarch/mpatan-fma.c | 10 ++++++
sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c | 9 +++++
sysdeps/x86_64/fpu/multiarch/mpexp-fma.c | 9 +++++
sysdeps/x86_64/fpu/multiarch/mplog-fma.c | 8 +++++
sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c | 8 +++++
sysdeps/x86_64/fpu/multiarch/mptan-fma.c | 7 ++++
sysdeps/x86_64/fpu/multiarch/s_atan-fma.c | 9 +++++
sysdeps/x86_64/fpu/multiarch/s_atan.c | 6 ++--
sysdeps/x86_64/fpu/multiarch/s_sin-fma.c | 11 ++++++
sysdeps/x86_64/fpu/multiarch/s_sin.c | 12 ++++---
sysdeps/x86_64/fpu/multiarch/s_tan-fma.c | 8 +++++
sysdeps/x86_64/fpu/multiarch/s_tan.c | 6 ++--
sysdeps/x86_64/fpu/multiarch/sincos32-fma.c | 15 ++++++++
sysdeps/x86_64/fpu/multiarch/slowexp-fma.c | 9 +++++
sysdeps/x86_64/fpu/multiarch/slowpow-fma.c | 11 ++++++
31 files changed, 322 insertions(+), 19 deletions(-)
create mode 100644 sysdeps/x86_64/fpu/multiarch/doasin-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/dosincos-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/e_asin-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/e_atan2-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/e_exp-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/e_log-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/e_pow-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/halfulp-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mpa-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mpatan-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mpexp-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mplog-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/mptan-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_atan-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_sin-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_tan-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/sincos32-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/slowexp-fma.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/slowpow-fma.c
2012-10-01 H.J. Lu <hongjiu.lu@intel.com>
[BZ #14649]
* sysdeps/x86_64/fpu/multiarch/doasin-fma.c: New file.
* sysdeps/x86_64/fpu/multiarch/dosincos-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_asin-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_atan2-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_exp-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_log-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_pow-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/halfulp-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mpa-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mpatan-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mpexp-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mplog-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/mptan-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_atan-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_sin-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_tan-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/sincos32-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/slowexp-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/slowpow-fma.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/Makefile (libm-sysdep_routines):
Add FMA optimized functions.
(CFLAGS-doasin-fma.c): New.
(CFLAGS-dosincos-fma.c): Likewise.
(CFLAGS-e_asin-fma.c): Likewise.
(CFLAGS-e_atan2-fma.c): Likewise.
(CFLAGS-e_exp-fma.c): Likewise.
(CFLAGS-e_log-fma.c): Likewise.
(CFLAGS-e_pow-fma.c): Likewise.
(CFLAGS-halfulp-fma.c): Likewise.
(CFLAGS-mpa-fma.c): Likewise.
(CFLAGS-mpatan-fma.c): Likewise.
(CFLAGS-mpatan2-fma.c): Likewise.
(CFLAGS-mpexp-fma.c): Likewise.
(CFLAGS-mplog-fma.c): Likewise.
(CFLAGS-mpsqrt-fma.c): Likewise.
(CFLAGS-mptan-fma.c): Likewise.
(CFLAGS-s_atan-fma.c): Likewise.
(CFLAGS-sincos32-fma.c): Likewise.
(CFLAGS-slowexp-fma.c): Likewise.
(CFLAGS-slowpow-fma.c): Likewise.
(CFLAGS-s_sin-fma.c): Likewise.
(CFLAGS-s_tan-fma.c): Likewise.
* sysdeps/x86_64/fpu/multiarch/e_asin.c: Support FMA.
* sysdeps/x86_64/fpu/multiarch/e_atan2.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_exp.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_log.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_pow.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_atan.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_sin.c: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_tan.c: Likewise.
diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
index 12b0526..f59d5ea 100644
--- a/sysdeps/x86_64/fpu/multiarch/Makefile
+++ b/sysdeps/x86_64/fpu/multiarch/Makefile
@@ -2,6 +2,37 @@ ifeq ($(subdir),math)
libm-sysdep_routines += s_floor-c s_ceil-c s_floorf-c s_ceilf-c \
s_rint-c s_rintf-c s_nearbyint-c s_nearbyintf-c
+ifeq (yes,$(config-cflags-avx))
+libm-sysdep_routines += e_exp-fma e_log-fma e_pow-fma s_atan-fma \
+ e_asin-fma e_atan2-fma s_sin-fma s_tan-fma \
+ mplog-fma mpa-fma slowexp-fma slowpow-fma \
+ sincos32-fma doasin-fma dosincos-fma \
+ halfulp-fma mpexp-fma \
+ mpatan2-fma mpatan-fma mpsqrt-fma mptan-fma
+
+CFLAGS-doasin-fma.c = -mfma
+CFLAGS-dosincos-fma.c = -mfma
+CFLAGS-e_asin-fma.c = -mfma
+CFLAGS-e_atan2-fma.c = -mfma
+CFLAGS-e_exp-fma.c = -mfma
+CFLAGS-e_log-fma.c = -mfma
+CFLAGS-e_pow-fma.c = -mfma
+CFLAGS-halfulp-fma.c = -mfma
+CFLAGS-mpa-fma.c = -mfma
+CFLAGS-mpatan-fma.c = -mfma
+CFLAGS-mpatan2-fma.c = -mfma
+CFLAGS-mpexp-fma.c = -mfma
+CFLAGS-mplog-fma.c = -mfma
+CFLAGS-mpsqrt-fma.c = -mfma
+CFLAGS-mptan-fma.c = -mfma
+CFLAGS-s_atan-fma.c = -mfma
+CFLAGS-sincos32-fma.c = -mfma
+CFLAGS-slowexp-fma.c = -mfma
+CFLAGS-slowpow-fma.c = -mfma
+CFLAGS-s_sin-fma.c = -mfma
+CFLAGS-s_tan-fma.c = -mfma
+endif
+
ifeq ($(have-mfma4),yes)
libm-sysdep_routines += e_exp-fma4 e_log-fma4 e_pow-fma4 s_atan-fma4 \
e_asin-fma4 e_atan2-fma4 s_sin-fma4 s_tan-fma4 \
diff --git a/sysdeps/x86_64/fpu/multiarch/doasin-fma.c b/sysdeps/x86_64/fpu/multiarch/doasin-fma.c
new file mode 100644
index 0000000..7a09865
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/doasin-fma.c
@@ -0,0 +1,4 @@
+#define __doasin __doasin_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/doasin.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/dosincos-fma.c b/sysdeps/x86_64/fpu/multiarch/dosincos-fma.c
new file mode 100644
index 0000000..5744586
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/dosincos-fma.c
@@ -0,0 +1,6 @@
+#define __docos __docos_fma
+#define __dubcos __dubcos_fma
+#define __dubsin __dubsin_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/dosincos.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_asin-fma.c b/sysdeps/x86_64/fpu/multiarch/e_asin-fma.c
new file mode 100644
index 0000000..50e9c64
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_asin-fma.c
@@ -0,0 +1,11 @@
+#define __ieee754_acos __ieee754_acos_fma
+#define __ieee754_asin __ieee754_asin_fma
+#define __cos32 __cos32_fma
+#define __doasin __doasin_fma
+#define __docos __docos_fma
+#define __dubcos __dubcos_fma
+#define __dubsin __dubsin_fma
+#define __sin32 __sin32_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 e742a9c..14e0859 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
+#ifdef HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math_private.h>
extern double __ieee754_acos_sse2 (double);
extern double __ieee754_asin_sse2 (double);
+extern double __ieee754_acos_fma (double);
+extern double __ieee754_asin_fma (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_FMA ? __ieee754_acos_fma
+ : (HAS_FMA4 ? __ieee754_acos_fma4 : __ieee754_acos_sse2));
strong_alias (__ieee754_acos, __acos_finite)
libm_ifunc (__ieee754_asin,
- HAS_FMA4 ? __ieee754_asin_fma4 : __ieee754_asin_sse2);
+ HAS_FMA ? __ieee754_asin_fma
+ : (HAS_FMA4 ? __ieee754_asin_fma4 : __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-fma.c b/sysdeps/x86_64/fpu/multiarch/e_atan2-fma.c
new file mode 100644
index 0000000..caba686
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_atan2-fma.c
@@ -0,0 +1,10 @@
+#define __ieee754_atan2 __ieee754_atan2_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __dvd __dvd_fma
+#define __mpatan2 __mpatan2_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 6867c6e..8b5bafe 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_atan2.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_atan2.c
@@ -4,6 +4,7 @@
extern double __ieee754_atan2_sse2 (double, double);
extern double __ieee754_atan2_avx (double, double);
+extern double __ieee754_atan2_fma (double, double);
# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_atan2_fma4 (double, double);
# else
@@ -13,8 +14,9 @@ extern double __ieee754_atan2_fma4 (double, double);
# endif
libm_ifunc (__ieee754_atan2,
- HAS_FMA4 ? __ieee754_atan2_fma4
- : (HAS_AVX ? __ieee754_atan2_avx : __ieee754_atan2_sse2));
+ HAS_FMA ? __ieee754_atan2_fma
+ : (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-fma.c b/sysdeps/x86_64/fpu/multiarch/e_exp-fma.c
new file mode 100644
index 0000000..6e0fdb7
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_exp-fma.c
@@ -0,0 +1,6 @@
+#define __ieee754_exp __ieee754_exp_fma
+#define __exp1 __exp1_fma
+#define __slowexp __slowexp_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 3c65028..5749da4 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_exp.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_exp.c
@@ -4,6 +4,7 @@
extern double __ieee754_exp_sse2 (double);
extern double __ieee754_exp_avx (double);
+extern double __ieee754_exp_fma (double);
# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_exp_fma4 (double);
# else
@@ -13,8 +14,9 @@ extern double __ieee754_exp_fma4 (double);
# endif
libm_ifunc (__ieee754_exp,
- HAS_FMA4 ? __ieee754_exp_fma4
- : (HAS_AVX ? __ieee754_exp_avx : __ieee754_exp_sse2));
+ HAS_FMA ? __ieee754_exp_fma
+ : (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-fma.c b/sysdeps/x86_64/fpu/multiarch/e_log-fma.c
new file mode 100644
index 0000000..a7123b1
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_log-fma.c
@@ -0,0 +1,8 @@
+#define __ieee754_log __ieee754_log_fma
+#define __mplog __mplog_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 05f3668..f243c3f 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_log.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_log.c
@@ -4,6 +4,7 @@
extern double __ieee754_log_sse2 (double);
extern double __ieee754_log_avx (double);
+extern double __ieee754_log_fma (double);
# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_log_fma4 (double);
# else
@@ -13,8 +14,9 @@ extern double __ieee754_log_fma4 (double);
# endif
libm_ifunc (__ieee754_log,
- HAS_FMA4 ? __ieee754_log_fma4
- : (HAS_AVX ? __ieee754_log_avx : __ieee754_log_sse2));
+ HAS_FMA ? __ieee754_log_fma
+ : (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/e_pow-fma.c b/sysdeps/x86_64/fpu/multiarch/e_pow-fma.c
new file mode 100644
index 0000000..6fd4083
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_pow-fma.c
@@ -0,0 +1,6 @@
+#define __ieee754_pow __ieee754_pow_fma
+#define __exp1 __exp1_fma
+#define __slowpow __slowpow_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/e_pow.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_pow.c b/sysdeps/x86_64/fpu/multiarch/e_pow.c
index a740b6c..f47fa95 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_pow.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_pow.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_pow_sse2 (double, double);
+extern double __ieee754_pow_fma (double, double);
+
+# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_pow_fma4 (double, double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __ieee754_pow_fma4 ((void *) 0)
+# endif
-libm_ifunc (__ieee754_pow, HAS_FMA4 ? __ieee754_pow_fma4 : __ieee754_pow_sse2);
+libm_ifunc (__ieee754_pow,
+ HAS_FMA ? __ieee754_pow_fma
+ : (HAS_FMA4 ? __ieee754_pow_fma4 : __ieee754_pow_sse2));
strong_alias (__ieee754_pow, __pow_finite)
# define __ieee754_pow __ieee754_pow_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/halfulp-fma.c b/sysdeps/x86_64/fpu/multiarch/halfulp-fma.c
new file mode 100644
index 0000000..6ca7046
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/halfulp-fma.c
@@ -0,0 +1,4 @@
+#define __halfulp __halfulp_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/halfulp.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpa-fma.c b/sysdeps/x86_64/fpu/multiarch/mpa-fma.c
new file mode 100644
index 0000000..5c182a7
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpa-fma.c
@@ -0,0 +1,12 @@
+#define __add __add_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define __dbl_mp __dbl_mp_fma
+#define __dvd __dvd_fma
+
+#define NO___CPY 1
+#define NO___MP_DBL 1
+#define NO___ACR 1
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mpa.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpatan-fma.c b/sysdeps/x86_64/fpu/multiarch/mpatan-fma.c
new file mode 100644
index 0000000..d216f91
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpatan-fma.c
@@ -0,0 +1,10 @@
+#define __mpatan __mpatan_fma
+#define __add __add_fma
+#define __dvd __dvd_fma
+#define __mpsqrt __mpsqrt_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define AVOID_MPATAN_H 1
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mpatan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c b/sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c
new file mode 100644
index 0000000..98df336
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpatan2-fma.c
@@ -0,0 +1,9 @@
+#define __mpatan2 __mpatan2_fma
+#define __add __add_fma
+#define __dvd __dvd_fma
+#define __mpatan __mpatan_fma
+#define __mpsqrt __mpsqrt_fma
+#define __mul __mul_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mpatan2.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpexp-fma.c b/sysdeps/x86_64/fpu/multiarch/mpexp-fma.c
new file mode 100644
index 0000000..637631b
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpexp-fma.c
@@ -0,0 +1,9 @@
+#define __mpexp __mpexp_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __dvd __dvd_fma
+#define __mul __mul_fma
+#define AVOID_MPEXP_H 1
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mpexp.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mplog-fma.c b/sysdeps/x86_64/fpu/multiarch/mplog-fma.c
new file mode 100644
index 0000000..645b6b7
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mplog-fma.c
@@ -0,0 +1,8 @@
+#define __mplog __mplog_fma
+#define __add __add_fma
+#define __mpexp __mpexp_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mplog.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c b/sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c
new file mode 100644
index 0000000..44d7a23
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpsqrt-fma.c
@@ -0,0 +1,8 @@
+#define __mpsqrt __mpsqrt_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define AVOID_MPSQRT_H 1
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mpsqrt.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mptan-fma.c b/sysdeps/x86_64/fpu/multiarch/mptan-fma.c
new file mode 100644
index 0000000..d1a6914
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mptan-fma.c
@@ -0,0 +1,7 @@
+#define __mptan __mptan_fma
+#define __c32 __c32_fma
+#define __dvd __dvd_fma
+#define __mpranred __mpranred_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/mptan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_atan-fma.c b/sysdeps/x86_64/fpu/multiarch/s_atan-fma.c
new file mode 100644
index 0000000..bedb3f2
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_atan-fma.c
@@ -0,0 +1,9 @@
+#define atan __atan_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mpatan __mpatan_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 ae16d7c..542e656 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_atan.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_atan.c
@@ -4,6 +4,7 @@
extern double __atan_sse2 (double);
extern double __atan_avx (double);
+extern double __atan_fma (double);
# ifdef HAVE_FMA4_SUPPORT
extern double __atan_fma4 (double);
# else
@@ -12,8 +13,9 @@ extern double __atan_fma4 (double);
# define __atan_fma4 ((void *) 0)
# endif
-libm_ifunc (atan, (HAS_FMA4 ? __atan_fma4 :
- HAS_AVX ? __atan_avx : __atan_sse2));
+libm_ifunc (atan, (HAS_FMA ? __atan_fma :
+ (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-fma.c b/sysdeps/x86_64/fpu/multiarch/s_sin-fma.c
new file mode 100644
index 0000000..15f3c39
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_sin-fma.c
@@ -0,0 +1,11 @@
+#define __cos __cos_fma
+#define __sin __sin_fma
+#define __docos __docos_fma
+#define __dubsin __dubsin_fma
+#define __mpcos __mpcos_fma
+#define __mpcos1 __mpcos1_fma
+#define __mpsin __mpsin_fma
+#define __mpsin1 __mpsin1_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 a0c2521..9405377 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_sin.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_sin.c
@@ -7,6 +7,8 @@ extern double __cos_sse2 (double);
extern double __sin_sse2 (double);
extern double __cos_avx (double);
extern double __sin_avx (double);
+extern double __cos_fma (double);
+extern double __sin_fma (double);
# ifdef HAVE_FMA4_SUPPORT
extern double __cos_fma4 (double);
extern double __sin_fma4 (double);
@@ -17,12 +19,14 @@ extern double __sin_fma4 (double);
# define __sin_fma4 ((void *) 0)
# endif
-libm_ifunc (__cos, (HAS_FMA4 ? __cos_fma4 :
- HAS_AVX ? __cos_avx : __cos_sse2));
+libm_ifunc (__cos, (HAS_FMA ? __cos_fma :
+ (HAS_FMA4 ? __cos_fma4 :
+ HAS_AVX ? __cos_avx : __cos_sse2)));
weak_alias (__cos, cos)
-libm_ifunc (__sin, (HAS_FMA4 ? __sin_fma4 :
- HAS_AVX ? __sin_avx : __sin_sse2));
+libm_ifunc (__sin, (HAS_FMA ? __sin_fma :
+ (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-fma.c b/sysdeps/x86_64/fpu/multiarch/s_tan-fma.c
new file mode 100644
index 0000000..c85f8bc
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_tan-fma.c
@@ -0,0 +1,8 @@
+#define tan __tan_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mpranred __mpranred_fma
+#define __mptan __mptan_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#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 904308f..4f25592 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_tan.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_tan.c
@@ -4,6 +4,7 @@
extern double __tan_sse2 (double);
extern double __tan_avx (double);
+extern double __tan_fma (double);
# ifdef HAVE_FMA4_SUPPORT
extern double __tan_fma4 (double);
# else
@@ -12,8 +13,9 @@ extern double __tan_fma4 (double);
# define __tan_fma4 ((void *) 0)
# endif
-libm_ifunc (tan, (HAS_FMA4 ? __tan_fma4 :
- HAS_AVX ? __tan_avx : __tan_sse2));
+libm_ifunc (tan, (HAS_FMA ? __tan_fma :
+ (HAS_FMA4 ? __tan_fma4 :
+ HAS_AVX ? __tan_avx : __tan_sse2)));
# define tan __tan_sse2
#endif
diff --git a/sysdeps/x86_64/fpu/multiarch/sincos32-fma.c b/sysdeps/x86_64/fpu/multiarch/sincos32-fma.c
new file mode 100644
index 0000000..dcd44bc
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/sincos32-fma.c
@@ -0,0 +1,15 @@
+#define __cos32 __cos32_fma
+#define __sin32 __sin32_fma
+#define __c32 __c32_fma
+#define __mpsin __mpsin_fma
+#define __mpsin1 __mpsin1_fma
+#define __mpcos __mpcos_fma
+#define __mpcos1 __mpcos1_fma
+#define __mpranred __mpranred_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/sincos32.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/slowexp-fma.c b/sysdeps/x86_64/fpu/multiarch/slowexp-fma.c
new file mode 100644
index 0000000..6fffca1
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/slowexp-fma.c
@@ -0,0 +1,9 @@
+#define __slowexp __slowexp_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mpexp __mpexp_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/slowexp.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/slowpow-fma.c b/sysdeps/x86_64/fpu/multiarch/slowpow-fma.c
new file mode 100644
index 0000000..160ed68
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/slowpow-fma.c
@@ -0,0 +1,11 @@
+#define __slowpow __slowpow_fma
+#define __add __add_fma
+#define __dbl_mp __dbl_mp_fma
+#define __mpexp __mpexp_fma
+#define __mplog __mplog_fma
+#define __mul __mul_fma
+#define __sub __sub_fma
+#define __halfulp __halfulp_fma
+#define SECTION __attribute__ ((section (".text.fma")))
+
+#include <sysdeps/ieee754/dbl-64/slowpow.c>
--
1.7.11.4