From e0016b11d6186a7003d7204cb100acab9bbcd940 Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Tue, 25 Oct 2011 21:34:55 -0400 Subject: Add AVX optimized versions for some x86-64 math functions --- sysdeps/x86_64/fpu/multiarch/Makefile | 33 +++++++++++++++++++++++++++-- sysdeps/x86_64/fpu/multiarch/brandred-avx.c | 4 ++++ sysdeps/x86_64/fpu/multiarch/doasin-avx.c | 4 ++++ sysdeps/x86_64/fpu/multiarch/dosincos-avx.c | 6 ++++++ sysdeps/x86_64/fpu/multiarch/e_asin-avx.c | 11 ++++++++++ sysdeps/x86_64/fpu/multiarch/e_asin.c | 19 +++++++++++++---- sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c | 10 +++++++++ sysdeps/x86_64/fpu/multiarch/e_atan2.c | 12 +++++++++-- sysdeps/x86_64/fpu/multiarch/e_exp-avx.c | 6 ++++++ sysdeps/x86_64/fpu/multiarch/e_exp.c | 13 ++++++++++-- sysdeps/x86_64/fpu/multiarch/e_log-avx.c | 8 +++++++ sysdeps/x86_64/fpu/multiarch/e_log.c | 14 ++++++++++-- sysdeps/x86_64/fpu/multiarch/mpa-avx.c | 12 +++++++++++ sysdeps/x86_64/fpu/multiarch/mpatan-avx.c | 10 +++++++++ sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c | 9 ++++++++ sysdeps/x86_64/fpu/multiarch/mpexp-avx.c | 9 ++++++++ sysdeps/x86_64/fpu/multiarch/mplog-avx.c | 8 +++++++ sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c | 8 +++++++ sysdeps/x86_64/fpu/multiarch/mptan-avx.c | 7 ++++++ sysdeps/x86_64/fpu/multiarch/s_atan-avx.c | 9 ++++++++ sysdeps/x86_64/fpu/multiarch/s_atan.c | 11 ++++++++-- sysdeps/x86_64/fpu/multiarch/s_sin-avx.c | 12 +++++++++++ sysdeps/x86_64/fpu/multiarch/s_sin.c | 17 +++++++++++---- sysdeps/x86_64/fpu/multiarch/s_tan-avx.c | 9 ++++++++ sysdeps/x86_64/fpu/multiarch/s_tan.c | 11 ++++++++-- sysdeps/x86_64/fpu/multiarch/sincos32-avx.c | 15 +++++++++++++ sysdeps/x86_64/fpu/multiarch/slowexp-avx.c | 9 ++++++++ 27 files changed, 276 insertions(+), 20 deletions(-) create mode 100644 sysdeps/x86_64/fpu/multiarch/brandred-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/doasin-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/dosincos-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/e_asin-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/e_exp-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/e_log-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mpa-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mpatan-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mpexp-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mplog-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/mptan-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/s_atan-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/s_sin-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/s_tan-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/sincos32-avx.c create mode 100644 sysdeps/x86_64/fpu/multiarch/slowexp-avx.c (limited to 'sysdeps/x86_64/fpu') 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 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 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 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 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 # include 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 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 # include 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 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 # include 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 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 # include 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 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 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 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 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 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 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 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 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 # include 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 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 # include # 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 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 # include 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 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 -- cgit 1.4.1