From: "H.J. Lu" <hjl.tools@gmail.com>
To: Sunil K Pandey <skpgkp2@gmail.com>
Cc: libc-alpha@sourceware.org, hjl.tools@gmail.com,
andrey.kolesov@intel.com, marius.cornea@intel.com
Subject: Re: [PATCH v5 10/18] x86-64: Add vector atan2/atan2f implementation to libmvec
Date: Wed, 29 Dec 2021 13:26:26 -0800 [thread overview]
Message-ID: <YczSgo6m2tvatLYO@gmail.com> (raw)
In-Reply-To: <20211229064000.1465621-11-skpgkp2@gmail.com>
On Tue, Dec 28, 2021 at 10:39:52PM -0800, Sunil K Pandey wrote:
> Implement vectorized atan2/atan2f containing SSE, AVX, AVX2 and
> AVX512 versions for libmvec as per vector ABI. It also contains
> accuracy and ABI tests for vector atan2/atan2f with regenerated ulps.
> ---
> bits/libm-simd-decl-stubs.h | 11 +
> math/bits/mathcalls.h | 2 +-
> .../unix/sysv/linux/x86_64/libmvec.abilist | 8 +
> sysdeps/x86/fpu/bits/math-vector.h | 4 +
> .../x86/fpu/finclude/math-vector-fortran.h | 4 +
> sysdeps/x86_64/fpu/Makeconfig | 1 +
> sysdeps/x86_64/fpu/Versions | 2 +
> sysdeps/x86_64/fpu/libm-test-ulps | 20 +
> .../fpu/multiarch/svml_d_atan22_core-sse2.S | 20 +
> .../x86_64/fpu/multiarch/svml_d_atan22_core.c | 28 ++
> .../fpu/multiarch/svml_d_atan22_core_sse4.S | 471 +++++++++++++++++
> .../fpu/multiarch/svml_d_atan24_core-sse.S | 20 +
> .../x86_64/fpu/multiarch/svml_d_atan24_core.c | 28 ++
> .../fpu/multiarch/svml_d_atan24_core_avx2.S | 451 +++++++++++++++++
> .../fpu/multiarch/svml_d_atan28_core-avx2.S | 20 +
> .../x86_64/fpu/multiarch/svml_d_atan28_core.c | 28 ++
> .../fpu/multiarch/svml_d_atan28_core_avx512.S | 475 ++++++++++++++++++
> .../fpu/multiarch/svml_s_atan2f16_core-avx2.S | 20 +
> .../fpu/multiarch/svml_s_atan2f16_core.c | 28 ++
> .../multiarch/svml_s_atan2f16_core_avx512.S | 399 +++++++++++++++
> .../fpu/multiarch/svml_s_atan2f4_core-sse2.S | 20 +
> .../fpu/multiarch/svml_s_atan2f4_core.c | 28 ++
> .../fpu/multiarch/svml_s_atan2f4_core_sse4.S | 384 ++++++++++++++
> .../fpu/multiarch/svml_s_atan2f8_core-sse.S | 20 +
> .../fpu/multiarch/svml_s_atan2f8_core.c | 28 ++
> .../fpu/multiarch/svml_s_atan2f8_core_avx2.S | 362 +++++++++++++
> sysdeps/x86_64/fpu/svml_d_atan22_core.S | 29 ++
> sysdeps/x86_64/fpu/svml_d_atan24_core.S | 29 ++
> sysdeps/x86_64/fpu/svml_d_atan24_core_avx.S | 25 +
> sysdeps/x86_64/fpu/svml_d_atan28_core.S | 25 +
> sysdeps/x86_64/fpu/svml_s_atan2f16_core.S | 25 +
> sysdeps/x86_64/fpu/svml_s_atan2f4_core.S | 29 ++
> sysdeps/x86_64/fpu/svml_s_atan2f8_core.S | 29 ++
> sysdeps/x86_64/fpu/svml_s_atan2f8_core_avx.S | 25 +
> .../fpu/test-double-libmvec-atan2-avx.c | 1 +
> .../fpu/test-double-libmvec-atan2-avx2.c | 1 +
> .../fpu/test-double-libmvec-atan2-avx512f.c | 1 +
> .../x86_64/fpu/test-double-libmvec-atan2.c | 3 +
> .../x86_64/fpu/test-double-vlen2-wrappers.c | 1 +
> .../fpu/test-double-vlen4-avx2-wrappers.c | 1 +
> .../x86_64/fpu/test-double-vlen4-wrappers.c | 1 +
> .../x86_64/fpu/test-double-vlen8-wrappers.c | 1 +
> .../fpu/test-float-libmvec-atan2f-avx.c | 1 +
> .../fpu/test-float-libmvec-atan2f-avx2.c | 1 +
> .../fpu/test-float-libmvec-atan2f-avx512f.c | 1 +
> .../x86_64/fpu/test-float-libmvec-atan2f.c | 3 +
> .../x86_64/fpu/test-float-vlen16-wrappers.c | 1 +
> .../x86_64/fpu/test-float-vlen4-wrappers.c | 1 +
> .../fpu/test-float-vlen8-avx2-wrappers.c | 1 +
> .../x86_64/fpu/test-float-vlen8-wrappers.c | 1 +
> 50 files changed, 3117 insertions(+), 1 deletion(-)
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core-sse2.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core_sse4.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core-sse.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core_avx2.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core-avx2.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core_avx512.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core-avx2.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core_avx512.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core-sse2.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core_sse4.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core-sse.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core_avx2.S
> create mode 100644 sysdeps/x86_64/fpu/svml_d_atan22_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_d_atan24_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_d_atan24_core_avx.S
> create mode 100644 sysdeps/x86_64/fpu/svml_d_atan28_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_s_atan2f16_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_s_atan2f4_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_s_atan2f8_core.S
> create mode 100644 sysdeps/x86_64/fpu/svml_s_atan2f8_core_avx.S
> create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx.c
> create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx2.c
> create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx512f.c
> create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-atan2.c
> create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx.c
> create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx2.c
> create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx512f.c
> create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-atan2f.c
>
> diff --git a/bits/libm-simd-decl-stubs.h b/bits/libm-simd-decl-stubs.h
> index 7f1304ed1d..31878bf4ed 100644
> --- a/bits/libm-simd-decl-stubs.h
> +++ b/bits/libm-simd-decl-stubs.h
> @@ -208,4 +208,15 @@
> #define __DECL_SIMD_cbrtf32x
> #define __DECL_SIMD_cbrtf64x
> #define __DECL_SIMD_cbrtf128x
> +
> +#define __DECL_SIMD_atan2
> +#define __DECL_SIMD_atan2f
> +#define __DECL_SIMD_atan2l
> +#define __DECL_SIMD_atan2f16
> +#define __DECL_SIMD_atan2f32
> +#define __DECL_SIMD_atan2f64
> +#define __DECL_SIMD_atan2f128
> +#define __DECL_SIMD_atan2f32x
> +#define __DECL_SIMD_atan2f64x
> +#define __DECL_SIMD_atan2f128x
> #endif
> diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h
> index 26d18f0135..1bd4911993 100644
> --- a/math/bits/mathcalls.h
> +++ b/math/bits/mathcalls.h
> @@ -56,7 +56,7 @@ __MATHCALL_VEC (asin,, (_Mdouble_ __x));
> /* Arc tangent of X. */
> __MATHCALL_VEC (atan,, (_Mdouble_ __x));
> /* Arc tangent of Y/X. */
> -__MATHCALL (atan2,, (_Mdouble_ __y, _Mdouble_ __x));
> +__MATHCALL_VEC (atan2,, (_Mdouble_ __y, _Mdouble_ __x));
>
> /* Cosine of X. */
> __MATHCALL_VEC (cos,, (_Mdouble_ __x));
> diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
> index a6558d9810..2b3b8d3886 100644
> --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
> +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
> @@ -55,6 +55,7 @@ GLIBC_2.35 _ZGVbN2v_exp10 F
> GLIBC_2.35 _ZGVbN2v_exp2 F
> GLIBC_2.35 _ZGVbN2v_expm1 F
> GLIBC_2.35 _ZGVbN2v_sinh F
> +GLIBC_2.35 _ZGVbN2vv_atan2 F
> GLIBC_2.35 _ZGVbN2vv_hypot F
> GLIBC_2.35 _ZGVbN4v_acosf F
> GLIBC_2.35 _ZGVbN4v_asinf F
> @@ -65,6 +66,7 @@ GLIBC_2.35 _ZGVbN4v_exp10f F
> GLIBC_2.35 _ZGVbN4v_exp2f F
> GLIBC_2.35 _ZGVbN4v_expm1f F
> GLIBC_2.35 _ZGVbN4v_sinhf F
> +GLIBC_2.35 _ZGVbN4vv_atan2f F
> GLIBC_2.35 _ZGVbN4vv_hypotf F
> GLIBC_2.35 _ZGVcN4v_acos F
> GLIBC_2.35 _ZGVcN4v_asin F
> @@ -75,6 +77,7 @@ GLIBC_2.35 _ZGVcN4v_exp10 F
> GLIBC_2.35 _ZGVcN4v_exp2 F
> GLIBC_2.35 _ZGVcN4v_expm1 F
> GLIBC_2.35 _ZGVcN4v_sinh F
> +GLIBC_2.35 _ZGVcN4vv_atan2 F
> GLIBC_2.35 _ZGVcN4vv_hypot F
> GLIBC_2.35 _ZGVcN8v_acosf F
> GLIBC_2.35 _ZGVcN8v_asinf F
> @@ -85,6 +88,7 @@ GLIBC_2.35 _ZGVcN8v_exp10f F
> GLIBC_2.35 _ZGVcN8v_exp2f F
> GLIBC_2.35 _ZGVcN8v_expm1f F
> GLIBC_2.35 _ZGVcN8v_sinhf F
> +GLIBC_2.35 _ZGVcN8vv_atan2f F
> GLIBC_2.35 _ZGVcN8vv_hypotf F
> GLIBC_2.35 _ZGVdN4v_acos F
> GLIBC_2.35 _ZGVdN4v_asin F
> @@ -95,6 +99,7 @@ GLIBC_2.35 _ZGVdN4v_exp10 F
> GLIBC_2.35 _ZGVdN4v_exp2 F
> GLIBC_2.35 _ZGVdN4v_expm1 F
> GLIBC_2.35 _ZGVdN4v_sinh F
> +GLIBC_2.35 _ZGVdN4vv_atan2 F
> GLIBC_2.35 _ZGVdN4vv_hypot F
> GLIBC_2.35 _ZGVdN8v_acosf F
> GLIBC_2.35 _ZGVdN8v_asinf F
> @@ -105,6 +110,7 @@ GLIBC_2.35 _ZGVdN8v_exp10f F
> GLIBC_2.35 _ZGVdN8v_exp2f F
> GLIBC_2.35 _ZGVdN8v_expm1f F
> GLIBC_2.35 _ZGVdN8v_sinhf F
> +GLIBC_2.35 _ZGVdN8vv_atan2f F
> GLIBC_2.35 _ZGVdN8vv_hypotf F
> GLIBC_2.35 _ZGVeN16v_acosf F
> GLIBC_2.35 _ZGVeN16v_asinf F
> @@ -115,6 +121,7 @@ GLIBC_2.35 _ZGVeN16v_exp10f F
> GLIBC_2.35 _ZGVeN16v_exp2f F
> GLIBC_2.35 _ZGVeN16v_expm1f F
> GLIBC_2.35 _ZGVeN16v_sinhf F
> +GLIBC_2.35 _ZGVeN16vv_atan2f F
> GLIBC_2.35 _ZGVeN16vv_hypotf F
> GLIBC_2.35 _ZGVeN8v_acos F
> GLIBC_2.35 _ZGVeN8v_asin F
> @@ -125,4 +132,5 @@ GLIBC_2.35 _ZGVeN8v_exp10 F
> GLIBC_2.35 _ZGVeN8v_exp2 F
> GLIBC_2.35 _ZGVeN8v_expm1 F
> GLIBC_2.35 _ZGVeN8v_sinh F
> +GLIBC_2.35 _ZGVeN8vv_atan2 F
> GLIBC_2.35 _ZGVeN8vv_hypot F
> diff --git a/sysdeps/x86/fpu/bits/math-vector.h b/sysdeps/x86/fpu/bits/math-vector.h
> index dcd45934ab..62f2890ab3 100644
> --- a/sysdeps/x86/fpu/bits/math-vector.h
> +++ b/sysdeps/x86/fpu/bits/math-vector.h
> @@ -98,6 +98,10 @@
> # define __DECL_SIMD_cbrt __DECL_SIMD_x86_64
> # undef __DECL_SIMD_cbrtf
> # define __DECL_SIMD_cbrtf __DECL_SIMD_x86_64
> +# undef __DECL_SIMD_atan2
> +# define __DECL_SIMD_atan2 __DECL_SIMD_x86_64
> +# undef __DECL_SIMD_atan2f
> +# define __DECL_SIMD_atan2f __DECL_SIMD_x86_64
>
> # endif
> #endif
> diff --git a/sysdeps/x86/fpu/finclude/math-vector-fortran.h b/sysdeps/x86/fpu/finclude/math-vector-fortran.h
> index dfb5f13ea3..2269b74d50 100644
> --- a/sysdeps/x86/fpu/finclude/math-vector-fortran.h
> +++ b/sysdeps/x86/fpu/finclude/math-vector-fortran.h
> @@ -48,6 +48,8 @@
> !GCC$ builtin (sinhf) attributes simd (notinbranch) if('x86_64')
> !GCC$ builtin (cbrt) attributes simd (notinbranch) if('x86_64')
> !GCC$ builtin (cbrtf) attributes simd (notinbranch) if('x86_64')
> +!GCC$ builtin (atan2) attributes simd (notinbranch) if('x86_64')
> +!GCC$ builtin (atan2f) attributes simd (notinbranch) if('x86_64')
>
> !GCC$ builtin (cos) attributes simd (notinbranch) if('x32')
> !GCC$ builtin (cosf) attributes simd (notinbranch) if('x32')
> @@ -81,3 +83,5 @@
> !GCC$ builtin (sinhf) attributes simd (notinbranch) if('x32')
> !GCC$ builtin (cbrt) attributes simd (notinbranch) if('x32')
> !GCC$ builtin (cbrtf) attributes simd (notinbranch) if('x32')
> +!GCC$ builtin (atan2) attributes simd (notinbranch) if('x32')
> +!GCC$ builtin (atan2f) attributes simd (notinbranch) if('x32')
> diff --git a/sysdeps/x86_64/fpu/Makeconfig b/sysdeps/x86_64/fpu/Makeconfig
> index dde737c0d6..96a40856fa 100644
> --- a/sysdeps/x86_64/fpu/Makeconfig
> +++ b/sysdeps/x86_64/fpu/Makeconfig
> @@ -25,6 +25,7 @@ libmvec-funcs = \
> acos \
> asin \
> atan \
> + atan2 \
> cbrt \
> cos \
> cosh \
> diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions
> index b70aeb3e2f..f58c98eb45 100644
> --- a/sysdeps/x86_64/fpu/Versions
> +++ b/sysdeps/x86_64/fpu/Versions
> @@ -23,6 +23,7 @@ libmvec {
> _ZGVbN2v_exp2; _ZGVcN4v_exp2; _ZGVdN4v_exp2; _ZGVeN8v_exp2;
> _ZGVbN2v_expm1; _ZGVcN4v_expm1; _ZGVdN4v_expm1; _ZGVeN8v_expm1;
> _ZGVbN2v_sinh; _ZGVcN4v_sinh; _ZGVdN4v_sinh; _ZGVeN8v_sinh;
> + _ZGVbN2vv_atan2; _ZGVcN4vv_atan2; _ZGVdN4vv_atan2; _ZGVeN8vv_atan2;
> _ZGVbN2vv_hypot; _ZGVcN4vv_hypot; _ZGVdN4vv_hypot; _ZGVeN8vv_hypot;
> _ZGVbN4v_acosf; _ZGVcN8v_acosf; _ZGVdN8v_acosf; _ZGVeN16v_acosf;
> _ZGVbN4v_asinf; _ZGVcN8v_asinf; _ZGVdN8v_asinf; _ZGVeN16v_asinf;
> @@ -33,6 +34,7 @@ libmvec {
> _ZGVbN4v_exp2f; _ZGVcN8v_exp2f; _ZGVdN8v_exp2f; _ZGVeN16v_exp2f;
> _ZGVbN4v_expm1f; _ZGVcN8v_expm1f; _ZGVdN8v_expm1f; _ZGVeN16v_expm1f;
> _ZGVbN4v_sinhf; _ZGVcN8v_sinhf; _ZGVdN8v_sinhf; _ZGVeN16v_sinhf;
> + _ZGVbN4vv_atan2f; _ZGVcN8vv_atan2f; _ZGVdN8vv_atan2f; _ZGVeN16vv_atan2f;
> _ZGVbN4vv_hypotf; _ZGVcN8vv_hypotf; _ZGVdN8vv_hypotf; _ZGVeN16vv_hypotf;
> }
> }
> diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps
> index e039a993df..6f59c61756 100644
> --- a/sysdeps/x86_64/fpu/libm-test-ulps
> +++ b/sysdeps/x86_64/fpu/libm-test-ulps
> @@ -166,6 +166,26 @@ float: 2
> float128: 2
> ldouble: 1
>
> +Function: "atan2_vlen16":
> +float: 2
> +
> +Function: "atan2_vlen2":
> +double: 1
> +
> +Function: "atan2_vlen4":
> +double: 1
> +float: 2
> +
> +Function: "atan2_vlen4_avx2":
> +double: 1
> +
> +Function: "atan2_vlen8":
> +double: 1
> +float: 2
> +
> +Function: "atan2_vlen8_avx2":
> +float: 2
> +
> Function: "atan_downward":
> double: 1
> float: 2
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core-sse2.S
> new file mode 100644
> index 0000000000..6c3ad05a6c
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core-sse2.S
> @@ -0,0 +1,20 @@
> +/* SSE2 version of vectorized atan2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVbN2vv_atan2 _ZGVbN2vv_atan2_sse2
> +#include "../svml_d_atan22_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core.c
> new file mode 100644
> index 0000000000..43f1ee7f33
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized atan2, vector length is 2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVbN2vv_atan2
> +#include "ifunc-mathvec-sse4_1.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVbN2vv_atan2, __GI__ZGVbN2vv_atan2,
> + __redirect__ZGVbN2vv_atan2)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core_sse4.S
> new file mode 100644
> index 0000000000..5c0d0fd17f
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan22_core_sse4.S
> @@ -0,0 +1,471 @@
> +/* Function atan2 vectorized with SSE4.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_datan2_data_internal
> + */
> +#define dPI 0
> +#define dPIO2 16
> +#define dA19 32
> +#define dA18 48
> +#define dA17 64
> +#define dA16 80
> +#define dA15 96
> +#define dA14 112
> +#define dA13 128
> +#define dA12 144
> +#define dA11 160
> +#define dA10 176
> +#define dA09 192
> +#define dA08 208
> +#define dA07 224
> +#define dA06 240
> +#define dA05 256
> +#define dA04 272
> +#define dA03 288
> +#define dA02 304
> +#define dA01 320
> +#define dA00 336
> +#define dSIGN_MASK 352
> +#define iCHK_WORK_SUB 368
> +#define iCHK_WORK_CMP 384
> +#define dABS_MASK 400
> +#define dZERO 416
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.sse4,"ax",@progbits
> +ENTRY(_ZGVbN2vv_atan2_sse4)
> + subq $88, %rsp
> + cfi_def_cfa_offset(96)
> + movaps %xmm0, %xmm8
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Get r0~=1/B
> + * Cannot be replaced by VQRCP(D, dR0, dB);
> + * Argument Absolute values
> + */
> + movups dABS_MASK+__svml_datan2_data_internal(%rip), %xmm4
> + movaps %xmm1, %xmm9
> + movaps %xmm4, %xmm1
> + andps %xmm8, %xmm4
> + andps %xmm9, %xmm1
> + movaps %xmm4, %xmm2
> + cmpnltpd %xmm1, %xmm2
> +
> +/* Argument signs */
> + movups dSIGN_MASK+__svml_datan2_data_internal(%rip), %xmm3
> + movaps %xmm2, %xmm0
> + movups dPIO2+__svml_datan2_data_internal(%rip), %xmm5
> + movaps %xmm3, %xmm7
> + movaps %xmm3, %xmm6
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + orps %xmm1, %xmm3
> + movaps %xmm2, %xmm10
> + andps %xmm2, %xmm5
> + andnps %xmm4, %xmm0
> + andps %xmm2, %xmm3
> + andnps %xmm1, %xmm10
> + andps %xmm4, %xmm2
> + orps %xmm3, %xmm0
> + orps %xmm2, %xmm10
> + divpd %xmm10, %xmm0
> + movq iCHK_WORK_SUB+__svml_datan2_data_internal(%rip), %xmm11
> +
> +/* if x<0, dPI = Pi, else dPI =0 */
> + movaps %xmm9, %xmm3
> +
> +/* Check if y and x are on main path. */
> + pshufd $221, %xmm1, %xmm12
> + andps %xmm9, %xmm7
> + psubd %xmm11, %xmm12
> + andps %xmm8, %xmm6
> + movq iCHK_WORK_CMP+__svml_datan2_data_internal(%rip), %xmm13
> + xorl %edx, %edx
> + movups %xmm4, 16(%rsp)
> + xorl %eax, %eax
> + pshufd $221, %xmm4, %xmm14
> + movdqa %xmm12, %xmm4
> + pcmpgtd %xmm13, %xmm4
> + pcmpeqd %xmm13, %xmm12
> + por %xmm12, %xmm4
> +
> +/* Polynomial. */
> + movaps %xmm0, %xmm12
> + mulpd %xmm0, %xmm12
> + cmplepd dZERO+__svml_datan2_data_internal(%rip), %xmm3
> + psubd %xmm11, %xmm14
> + movdqa %xmm14, %xmm15
> + pcmpeqd %xmm13, %xmm14
> + pcmpgtd %xmm13, %xmm15
> + por %xmm14, %xmm15
> + movaps %xmm12, %xmm14
> + mulpd %xmm12, %xmm14
> + por %xmm15, %xmm4
> + movaps %xmm14, %xmm15
> + mulpd %xmm14, %xmm15
> + movmskps %xmm4, %ecx
> + movups %xmm10, (%rsp)
> + movups dA19+__svml_datan2_data_internal(%rip), %xmm10
> + mulpd %xmm15, %xmm10
> + movups dA18+__svml_datan2_data_internal(%rip), %xmm13
> + movups dA17+__svml_datan2_data_internal(%rip), %xmm11
> + addpd dA15+__svml_datan2_data_internal(%rip), %xmm10
> + mulpd %xmm15, %xmm13
> + mulpd %xmm15, %xmm11
> + mulpd %xmm15, %xmm10
> + addpd dA14+__svml_datan2_data_internal(%rip), %xmm13
> + addpd dA13+__svml_datan2_data_internal(%rip), %xmm11
> + addpd dA11+__svml_datan2_data_internal(%rip), %xmm10
> + mulpd %xmm15, %xmm13
> + mulpd %xmm15, %xmm11
> + mulpd %xmm15, %xmm10
> + addpd dA10+__svml_datan2_data_internal(%rip), %xmm13
> + addpd dA09+__svml_datan2_data_internal(%rip), %xmm11
> + addpd dA07+__svml_datan2_data_internal(%rip), %xmm10
> + mulpd %xmm15, %xmm13
> + mulpd %xmm15, %xmm11
> + mulpd %xmm15, %xmm10
> + addpd dA06+__svml_datan2_data_internal(%rip), %xmm13
> + addpd dA05+__svml_datan2_data_internal(%rip), %xmm11
> + addpd dA03+__svml_datan2_data_internal(%rip), %xmm10
> + mulpd %xmm15, %xmm13
> + mulpd %xmm15, %xmm11
> + mulpd %xmm12, %xmm10
> + addpd dA02+__svml_datan2_data_internal(%rip), %xmm13
> + addpd dA01+__svml_datan2_data_internal(%rip), %xmm11
> + addpd %xmm10, %xmm13
> + mulpd %xmm11, %xmm12
> + mulpd %xmm13, %xmm14
> + movups dA16+__svml_datan2_data_internal(%rip), %xmm2
> + mulpd %xmm15, %xmm2
> + addpd dA12+__svml_datan2_data_internal(%rip), %xmm2
> + mulpd %xmm15, %xmm2
> + addpd dA08+__svml_datan2_data_internal(%rip), %xmm2
> + mulpd %xmm15, %xmm2
> + addpd dA04+__svml_datan2_data_internal(%rip), %xmm2
> +
> +/* A00=1.0, account for it later VQFMA(D, dP4, dP4, dR8, dA00); */
> + mulpd %xmm2, %xmm15
> + addpd %xmm12, %xmm15
> + addpd %xmm14, %xmm15
> +
> +/*
> + * Reconstruction.
> + * dP=(R+R*dP) + dPIO2
> + */
> + mulpd %xmm0, %xmm15
> + addpd %xmm15, %xmm0
> + addpd %xmm5, %xmm0
> + andps __svml_datan2_data_internal(%rip), %xmm3
> + orps %xmm7, %xmm0
> + addpd %xmm3, %xmm0
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + movups 16(%rsp), %xmm11
> + orps %xmm6, %xmm0
> + testb $3, %cl
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm1 xmm3 xmm4 xmm5 xmm6 xmm7 xmm8 xmm9 xmm11
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm8 xmm9
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + addq $88, %rsp
> + cfi_def_cfa_offset(8)
> + ret
> + cfi_def_cfa_offset(96)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + movups %xmm8, 32(%rsp)
> + movups %xmm9, 48(%rsp)
> + movups %xmm0, 64(%rsp)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0
> +
> + movq %r12, 16(%rsp)
> + cfi_offset(12, -80)
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + cfi_offset(13, -88)
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + cfi_offset(14, -96)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $2, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx rbp r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + movups 64(%rsp), %xmm0
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + cfi_offset(12, -80)
> + cfi_offset(13, -88)
> + cfi_offset(14, -96)
> + # LOE rbx rbp r12 r13 r14 r15 xmm0
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movsd 32(%rsp,%r14,8), %xmm0
> + movsd 48(%rsp,%r14,8), %xmm1
> + call atan2@PLT
> + # LOE rbx rbp r14 r15 r12d r13d xmm0
> +
> + movsd %xmm0, 64(%rsp,%r14,8)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + movups dZERO+__svml_datan2_data_internal(%rip), %xmm2
> +
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + movaps %xmm9, %xmm12
> + movaps %xmm8, %xmm10
> + cmpordpd %xmm9, %xmm12
> + cmpordpd %xmm8, %xmm10
> + cmpeqpd %xmm2, %xmm1
> + cmpeqpd %xmm2, %xmm11
> + andps %xmm10, %xmm12
> + orps %xmm11, %xmm1
> + pshufd $221, %xmm1, %xmm1
> + pshufd $221, %xmm12, %xmm11
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + pand %xmm11, %xmm1
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + movdqa %xmm1, %xmm13
> + pandn %xmm4, %xmm13
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + movups (%rsp), %xmm4
> + cmpeqpd %xmm2, %xmm4
> +
> +/* Go to callout */
> + movmskps %xmm13, %edx
> +
> +/* Set sPIO2 to zero if den. is zero */
> + movaps %xmm4, %xmm15
> + andps %xmm2, %xmm4
> + andnps %xmm5, %xmm15
> + andl $3, %edx
> + orps %xmm4, %xmm15
> + pshufd $221, %xmm9, %xmm5
> + orps %xmm7, %xmm15
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + pshufd $221, %xmm2, %xmm7
> + pcmpgtd %xmm5, %xmm7
> + pshufd $80, %xmm7, %xmm14
> + andps %xmm3, %xmm14
> + addpd %xmm14, %xmm15
> +
> +/* Merge results from main and spec path */
> + pshufd $80, %xmm1, %xmm3
> + orps %xmm6, %xmm15
> + movdqa %xmm3, %xmm6
> + andps %xmm3, %xmm15
> + andnps %xmm0, %xmm6
> + movaps %xmm6, %xmm0
> + orps %xmm15, %xmm0
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm8 xmm9
> +END(_ZGVbN2vv_atan2_sse4)
> +
> + .section .rodata, "a"
> + .align 16
> +
> +#ifdef __svml_datan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(16)) VUINT32 dPI[2][2];
> + __declspec(align(16)) VUINT32 dPIO2[2][2];
> + __declspec(align(16)) VUINT32 dA19[2][2];
> + __declspec(align(16)) VUINT32 dA18[2][2];
> + __declspec(align(16)) VUINT32 dA17[2][2];
> + __declspec(align(16)) VUINT32 dA16[2][2];
> + __declspec(align(16)) VUINT32 dA15[2][2];
> + __declspec(align(16)) VUINT32 dA14[2][2];
> + __declspec(align(16)) VUINT32 dA13[2][2];
> + __declspec(align(16)) VUINT32 dA12[2][2];
> + __declspec(align(16)) VUINT32 dA11[2][2];
> + __declspec(align(16)) VUINT32 dA10[2][2];
> + __declspec(align(16)) VUINT32 dA09[2][2];
> + __declspec(align(16)) VUINT32 dA08[2][2];
> + __declspec(align(16)) VUINT32 dA07[2][2];
> + __declspec(align(16)) VUINT32 dA06[2][2];
> + __declspec(align(16)) VUINT32 dA05[2][2];
> + __declspec(align(16)) VUINT32 dA04[2][2];
> + __declspec(align(16)) VUINT32 dA03[2][2];
> + __declspec(align(16)) VUINT32 dA02[2][2];
> + __declspec(align(16)) VUINT32 dA01[2][2];
> + __declspec(align(16)) VUINT32 dA00[2][2];
> + __declspec(align(16)) VUINT32 dSIGN_MASK[2][2];
> + __declspec(align(16)) VUINT32 iCHK_WORK_SUB[4][1];
> + __declspec(align(16)) VUINT32 iCHK_WORK_CMP[4][1];
> + __declspec(align(16)) VUINT32 dABS_MASK[2][2];
> + __declspec(align(16)) VUINT32 dZERO[2][2];
> +} __svml_datan2_data_internal;
> +#endif
> +__svml_datan2_data_internal:
> + .quad 0x400921FB54442D18, 0x400921FB54442D18 //dPI
> + .align 16
> + .quad 0x3FF921FB54442D18, 0x3FF921FB54442D18 //dPIO2
> + .align 16
> + .quad 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3 // dA19
> + .align 16
> + .quad 0x3F2CED0A36665209, 0x3F2CED0A36665209 // dA18
> + .align 16
> + .quad 0xBF52E67C93954C23, 0xBF52E67C93954C23 // dA17
> + .align 16
> + .quad 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3 // dA16
> + .align 16
> + .quad 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD // dA15
> + .align 16
> + .quad 0x3F914F4C661116A5, 0x3F914F4C661116A5 // dA14
> + .align 16
> + .quad 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C // dA13
> + .align 16
> + .quad 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F // dA12
> + .align 16
> + .quad 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC // dA11
> + .align 16
> + .quad 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B // dA10
> + .align 16
> + .quad 0xBFAAD261EAA09954, 0xBFAAD261EAA09954 // dA09
> + .align 16
> + .quad 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF // dA08
> + .align 16
> + .quad 0xBFB11084009435E0, 0xBFB11084009435E0 // dA07
> + .align 16
> + .quad 0x3FB3B12A49295651, 0x3FB3B12A49295651 // dA06
> + .align 16
> + .quad 0xBFB745D009BADA94, 0xBFB745D009BADA94 // dA05
> + .align 16
> + .quad 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5 // dA04
> + .align 16
> + .quad 0xBFC2492491EE55C7, 0xBFC2492491EE55C7 // dA03
> + .align 16
> + .quad 0x3FC999999997EE34, 0x3FC999999997EE34 // dA02
> + .align 16
> + .quad 0xBFD55555555553C5, 0xBFD55555555553C5 // dA01
> + .align 16
> + .quad 0x3FF0000000000000, 0x3FF0000000000000 // dA00
> + .align 16
> + .quad 0x8000000000000000, 0x8000000000000000 //dSIGN_MASK
> + .align 16
> + .long 0x80300000, 0x80300000, 0x80300000, 0x80300000 //iCHK_WORK_SUB
> + .align 16
> + .long 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000 //iCHK_WORK_CMP
> + .align 16
> + .quad 0x7fffffffffffffff, 0x7fffffffffffffff //dABS_MASK
> + .align 16
> + .quad 0x0000000000000000, 0x0000000000000000 //dZERO
> + .align 16
> + .type __svml_datan2_data_internal,@object
> + .size __svml_datan2_data_internal,.-__svml_datan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core-sse.S
> new file mode 100644
> index 0000000000..0db843a088
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core-sse.S
> @@ -0,0 +1,20 @@
> +/* SSE version of vectorized atan2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVdN4vv_atan2 _ZGVdN4vv_atan2_sse_wrapper
> +#include "../svml_d_atan24_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core.c
> new file mode 100644
> index 0000000000..c2e2611584
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized atan2, vector length is 4.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVdN4vv_atan2
> +#include "ifunc-mathvec-avx2.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVdN4vv_atan2, __GI__ZGVdN4vv_atan2,
> + __redirect__ZGVdN4vv_atan2)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core_avx2.S
> new file mode 100644
> index 0000000000..cdf780715b
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan24_core_avx2.S
> @@ -0,0 +1,451 @@
> +/* Function atan2 vectorized with AVX2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_datan2_data_internal
> + */
> +#define dPI 0
> +#define dPIO2 32
> +#define dA19 64
> +#define dA18 96
> +#define dA17 128
> +#define dA16 160
> +#define dA15 192
> +#define dA14 224
> +#define dA13 256
> +#define dA12 288
> +#define dA11 320
> +#define dA10 352
> +#define dA09 384
> +#define dA08 416
> +#define dA07 448
> +#define dA06 480
> +#define dA05 512
> +#define dA04 544
> +#define dA03 576
> +#define dA02 608
> +#define dA01 640
> +#define dA00 672
> +#define dSIGN_MASK 704
> +#define iCHK_WORK_SUB 736
> +#define iCHK_WORK_CMP 768
> +#define dABS_MASK 800
> +#define dZERO 832
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.avx2,"ax",@progbits
> +ENTRY(_ZGVdN4vv_atan2_avx2)
> + pushq %rbp
> + cfi_def_cfa_offset(16)
> + movq %rsp, %rbp
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> + andq $-32, %rsp
> + subq $128, %rsp
> + xorl %edx, %edx
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Get r0~=1/B
> + * Cannot be replaced by VQRCP(D, dR0, dB);
> + * Argument Absolute values
> + */
> + vmovupd dABS_MASK+__svml_datan2_data_internal(%rip), %ymm5
> +
> +/* Argument signs */
> + vmovupd dSIGN_MASK+__svml_datan2_data_internal(%rip), %ymm4
> + vmovups iCHK_WORK_SUB+__svml_datan2_data_internal(%rip), %xmm13
> + vmovupd %ymm0, (%rsp)
> + vmovapd %ymm1, %ymm8
> + vandpd %ymm5, %ymm8, %ymm2
> + vandpd %ymm5, %ymm0, %ymm1
> + vcmpnlt_uqpd %ymm2, %ymm1, %ymm15
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + vorpd %ymm4, %ymm2, %ymm6
> + vblendvpd %ymm15, %ymm6, %ymm1, %ymm3
> + vblendvpd %ymm15, %ymm1, %ymm2, %ymm6
> + vdivpd %ymm6, %ymm3, %ymm14
> + vmovups iCHK_WORK_CMP+__svml_datan2_data_internal(%rip), %xmm3
> + vmovupd %ymm6, 32(%rsp)
> + vandpd %ymm4, %ymm0, %ymm7
> + vandpd %ymm4, %ymm8, %ymm5
> + vandpd dPIO2+__svml_datan2_data_internal(%rip), %ymm15, %ymm4
> +
> +/* Check if y and x are on main path. */
> + vextractf128 $1, %ymm2, %xmm9
> + vextractf128 $1, %ymm1, %xmm10
> + vshufps $221, %xmm9, %xmm2, %xmm11
> + vshufps $221, %xmm10, %xmm1, %xmm12
> + vpsubd %xmm13, %xmm11, %xmm0
> + vpsubd %xmm13, %xmm12, %xmm9
> + vpcmpgtd %xmm3, %xmm0, %xmm15
> + vpcmpeqd %xmm3, %xmm0, %xmm6
> + vpcmpgtd %xmm3, %xmm9, %xmm10
> + vpcmpeqd %xmm3, %xmm9, %xmm3
> + vpor %xmm6, %xmm15, %xmm11
> + vpor %xmm3, %xmm10, %xmm12
> +
> +/* Polynomial. */
> + vmulpd %ymm14, %ymm14, %ymm10
> + vpor %xmm12, %xmm11, %xmm3
> + vmovupd dA18+__svml_datan2_data_internal(%rip), %ymm9
> + vmovupd dA17+__svml_datan2_data_internal(%rip), %ymm12
> + vmovupd dA16+__svml_datan2_data_internal(%rip), %ymm15
> + vmulpd %ymm10, %ymm10, %ymm11
> +
> +/* if x<0, dPI = Pi, else dPI =0 */
> + vcmple_oqpd dZERO+__svml_datan2_data_internal(%rip), %ymm8, %ymm13
> + vmovmskps %xmm3, %eax
> + vmulpd %ymm11, %ymm11, %ymm0
> + vandpd __svml_datan2_data_internal(%rip), %ymm13, %ymm6
> + vmovupd dA19+__svml_datan2_data_internal(%rip), %ymm13
> + vfmadd213pd dA14+__svml_datan2_data_internal(%rip), %ymm0, %ymm9
> + vfmadd213pd dA13+__svml_datan2_data_internal(%rip), %ymm0, %ymm12
> + vfmadd213pd dA12+__svml_datan2_data_internal(%rip), %ymm0, %ymm15
> + vfmadd213pd dA15+__svml_datan2_data_internal(%rip), %ymm0, %ymm13
> + vfmadd213pd dA10+__svml_datan2_data_internal(%rip), %ymm0, %ymm9
> + vfmadd213pd dA09+__svml_datan2_data_internal(%rip), %ymm0, %ymm12
> + vfmadd213pd dA08+__svml_datan2_data_internal(%rip), %ymm0, %ymm15
> + vfmadd213pd dA11+__svml_datan2_data_internal(%rip), %ymm0, %ymm13
> + vfmadd213pd dA06+__svml_datan2_data_internal(%rip), %ymm0, %ymm9
> + vfmadd213pd dA05+__svml_datan2_data_internal(%rip), %ymm0, %ymm12
> + vfmadd213pd dA04+__svml_datan2_data_internal(%rip), %ymm0, %ymm15
> + vfmadd213pd dA07+__svml_datan2_data_internal(%rip), %ymm0, %ymm13
> + vfmadd213pd dA02+__svml_datan2_data_internal(%rip), %ymm0, %ymm9
> + vfmadd213pd dA01+__svml_datan2_data_internal(%rip), %ymm0, %ymm12
> + vfmadd213pd dA03+__svml_datan2_data_internal(%rip), %ymm0, %ymm13
> +
> +/* A00=1.0, account for it later VQFMA(D, dP4, dP4, dR8, dA00); */
> + vmulpd %ymm15, %ymm0, %ymm0
> + vfmadd213pd %ymm9, %ymm10, %ymm13
> + vfmadd213pd %ymm0, %ymm10, %ymm12
> + vfmadd213pd %ymm12, %ymm11, %ymm13
> +
> +/*
> + * Reconstruction.
> + * dP=(R+R*dP) + dPIO2
> + */
> + vfmadd213pd %ymm14, %ymm14, %ymm13
> + vaddpd %ymm13, %ymm4, %ymm14
> + vorpd %ymm5, %ymm14, %ymm0
> + vaddpd %ymm0, %ymm6, %ymm9
> + vorpd %ymm7, %ymm9, %ymm0
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + testl %eax, %eax
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx xmm3 ymm0 ymm1 ymm2 ymm4 ymm5 ymm6 ymm7 ymm8
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm8
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + movq %rbp, %rsp
> + popq %rbp
> + cfi_def_cfa(7, 8)
> + cfi_restore(6)
> + ret
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + vmovupd (%rsp), %ymm1
> + vmovupd %ymm8, 64(%rsp)
> + vmovupd %ymm0, 96(%rsp)
> + vmovupd %ymm1, 32(%rsp)
> + # LOE rbx r12 r13 r14 r15 edx ymm0
> +
> + xorl %eax, %eax
> + # LOE rbx r12 r13 r14 r15 eax edx
> +
> + vzeroupper
> + movq %r12, 16(%rsp)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $4, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + vmovupd 96(%rsp), %ymm0
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r12 r13 r14 r15 ymm0
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movsd 32(%rsp,%r14,8), %xmm0
> + movsd 64(%rsp,%r14,8), %xmm1
> + call atan2@PLT
> + # LOE rbx r14 r15 r12d r13d xmm0
> +
> + movsd %xmm0, 96(%rsp,%r14,8)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> + vmovupd (%rsp), %ymm11
> +
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + vmovupd dZERO+__svml_datan2_data_internal(%rip), %ymm10
> +
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + vcmpordpd %ymm8, %ymm8, %ymm12
> + vcmpordpd %ymm11, %ymm11, %ymm13
> + vcmpeqpd %ymm10, %ymm2, %ymm2
> + vcmpeqpd %ymm10, %ymm1, %ymm1
> + vandpd %ymm13, %ymm12, %ymm14
> + vorpd %ymm1, %ymm2, %ymm2
> + vextractf128 $1, %ymm14, %xmm15
> + vextractf128 $1, %ymm2, %xmm11
> + vshufps $221, %xmm15, %xmm14, %xmm9
> + vshufps $221, %xmm11, %xmm2, %xmm12
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + vcmpeqpd 32(%rsp), %ymm10, %ymm2
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + vpand %xmm9, %xmm12, %xmm1
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + vpandn %xmm3, %xmm1, %xmm3
> +
> +/* Go to callout */
> + vmovmskps %xmm3, %edx
> +
> +/* Set sPIO2 to zero if den. is zero */
> + vblendvpd %ymm2, %ymm10, %ymm4, %ymm4
> + vorpd %ymm5, %ymm4, %ymm5
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + vextractf128 $1, %ymm10, %xmm2
> + vextractf128 $1, %ymm8, %xmm3
> + vshufps $221, %xmm2, %xmm10, %xmm4
> + vshufps $221, %xmm3, %xmm8, %xmm9
> + vpcmpgtd %xmm9, %xmm4, %xmm12
> + vpshufd $80, %xmm12, %xmm11
> + vpshufd $250, %xmm12, %xmm13
> + vinsertf128 $1, %xmm13, %ymm11, %ymm14
> + vandpd %ymm6, %ymm14, %ymm6
> + vaddpd %ymm6, %ymm5, %ymm2
> + vorpd %ymm7, %ymm2, %ymm2
> +
> +/* Merge results from main and spec path */
> + vpshufd $80, %xmm1, %xmm7
> + vpshufd $250, %xmm1, %xmm1
> + vinsertf128 $1, %xmm1, %ymm7, %ymm3
> + vblendvpd %ymm3, %ymm2, %ymm0, %ymm0
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm8
> +END(_ZGVdN4vv_atan2_avx2)
> +
> + .section .rodata, "a"
> + .align 32
> +
> +#ifdef __svml_datan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(32)) VUINT32 dPI[4][2];
> + __declspec(align(32)) VUINT32 dPIO2[4][2];
> + __declspec(align(32)) VUINT32 dA19[4][2];
> + __declspec(align(32)) VUINT32 dA18[4][2];
> + __declspec(align(32)) VUINT32 dA17[4][2];
> + __declspec(align(32)) VUINT32 dA16[4][2];
> + __declspec(align(32)) VUINT32 dA15[4][2];
> + __declspec(align(32)) VUINT32 dA14[4][2];
> + __declspec(align(32)) VUINT32 dA13[4][2];
> + __declspec(align(32)) VUINT32 dA12[4][2];
> + __declspec(align(32)) VUINT32 dA11[4][2];
> + __declspec(align(32)) VUINT32 dA10[4][2];
> + __declspec(align(32)) VUINT32 dA09[4][2];
> + __declspec(align(32)) VUINT32 dA08[4][2];
> + __declspec(align(32)) VUINT32 dA07[4][2];
> + __declspec(align(32)) VUINT32 dA06[4][2];
> + __declspec(align(32)) VUINT32 dA05[4][2];
> + __declspec(align(32)) VUINT32 dA04[4][2];
> + __declspec(align(32)) VUINT32 dA03[4][2];
> + __declspec(align(32)) VUINT32 dA02[4][2];
> + __declspec(align(32)) VUINT32 dA01[4][2];
> + __declspec(align(32)) VUINT32 dA00[4][2];
> + __declspec(align(32)) VUINT32 dSIGN_MASK[4][2];
> + __declspec(align(32)) VUINT32 iCHK_WORK_SUB[8][1];
> + __declspec(align(32)) VUINT32 iCHK_WORK_CMP[8][1];
> + __declspec(align(32)) VUINT32 dABS_MASK[4][2];
> + __declspec(align(32)) VUINT32 dZERO[4][2];
> +} __svml_datan2_data_internal;
> +#endif
> +__svml_datan2_data_internal:
> + .quad 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18 //dPI
> + .align 32
> + .quad 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18 //dPIO2
> + .align 32
> + .quad 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3 // dA19
> + .align 32
> + .quad 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209 // dA18
> + .align 32
> + .quad 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23 // dA17
> + .align 32
> + .quad 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3 // dA16
> + .align 32
> + .quad 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD // dA15
> + .align 32
> + .quad 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5 // dA14
> + .align 32
> + .quad 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C // dA13
> + .align 32
> + .quad 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F // dA12
> + .align 32
> + .quad 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC // dA11
> + .align 32
> + .quad 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B // dA10
> + .align 32
> + .quad 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954 // dA09
> + .align 32
> + .quad 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF // dA08
> + .align 32
> + .quad 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0 // dA07
> + .align 32
> + .quad 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651 // dA06
> + .align 32
> + .quad 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94 // dA05
> + .align 32
> + .quad 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5 // dA04
> + .align 32
> + .quad 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7 // dA03
> + .align 32
> + .quad 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34 // dA02
> + .align 32
> + .quad 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5 // dA01
> + .align 32
> + .quad 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000 // dA00
> + .align 32
> + .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 //dSIGN_MASK
> + .align 32
> + .long 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000 //iCHK_WORK_SUB
> + .align 32
> + .long 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000 //iCHK_WORK_CMP
> + .align 32
> + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff //dABS_MASK
> + .align 32
> + .quad 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000 //dZERO
> + .align 32
> + .type __svml_datan2_data_internal,@object
> + .size __svml_datan2_data_internal,.-__svml_datan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core-avx2.S
> new file mode 100644
> index 0000000000..a8d34a6143
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core-avx2.S
> @@ -0,0 +1,20 @@
> +/* AVX2 version of vectorized atan2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVeN8vv_atan2 _ZGVeN8vv_atan2_avx2_wrapper
> +#include "../svml_d_atan28_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core.c
> new file mode 100644
> index 0000000000..a0897e9cf0
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized atan2, vector length is 8.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVeN8vv_atan2
> +#include "ifunc-mathvec-avx512-skx.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVeN8vv_atan2, __GI__ZGVeN8vv_atan2,
> + __redirect__ZGVeN8vv_atan2)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core_avx512.S
> new file mode 100644
> index 0000000000..6d18f5f757
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan28_core_avx512.S
> @@ -0,0 +1,475 @@
> +/* Function atan2 vectorized with AVX-512.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_datan2_data_internal
> + */
> +#define dPI 0
> +#define dPIO2 64
> +#define dA19 128
> +#define dA18 192
> +#define dA17 256
> +#define dA16 320
> +#define dA15 384
> +#define dA14 448
> +#define dA13 512
> +#define dA12 576
> +#define dA11 640
> +#define dA10 704
> +#define dA09 768
> +#define dA08 832
> +#define dA07 896
> +#define dA06 960
> +#define dA05 1024
> +#define dA04 1088
> +#define dA03 1152
> +#define dA02 1216
> +#define dA01 1280
> +#define dA00 1344
> +#define dSIGN_MASK 1408
> +#define iCHK_WORK_SUB 1472
> +#define iCHK_WORK_CMP 1536
> +#define dABS_MASK 1600
> +#define dZERO 1664
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.evex512,"ax",@progbits
> +ENTRY(_ZGVeN8vv_atan2_skx)
> + pushq %rbp
> + cfi_def_cfa_offset(16)
> + movq %rsp, %rbp
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> + andq $-64, %rsp
> + subq $256, %rsp
> + xorl %edx, %edx
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Get r0~=1/B
> + * Cannot be replaced by VQRCP(D, dR0, dB);
> + * Argument Absolute values
> + */
> + vmovups dABS_MASK+__svml_datan2_data_internal(%rip), %zmm4
> +
> +/* Argument signs */
> + vmovups dSIGN_MASK+__svml_datan2_data_internal(%rip), %zmm6
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + vmovups dPIO2+__svml_datan2_data_internal(%rip), %zmm3
> + vandpd %zmm4, %zmm0, %zmm11
> + vmovaps %zmm1, %zmm7
> + vandpd %zmm4, %zmm7, %zmm2
> + vandpd %zmm6, %zmm7, %zmm5
> + vandpd %zmm6, %zmm0, %zmm4
> + vorpd %zmm6, %zmm2, %zmm12
> + vcmppd $17, {sae}, %zmm2, %zmm11, %k1
> + vmovdqu iCHK_WORK_CMP+__svml_datan2_data_internal(%rip), %ymm6
> + vmovups %zmm11, 64(%rsp)
> +
> +/* Check if y and x are on main path. */
> + vpsrlq $32, %zmm2, %zmm9
> + vblendmpd %zmm11, %zmm12, %zmm13{%k1}
> + vblendmpd %zmm2, %zmm11, %zmm15{%k1}
> + vpsrlq $32, %zmm11, %zmm8
> + vmovdqu iCHK_WORK_SUB+__svml_datan2_data_internal(%rip), %ymm12
> + vdivpd {rn-sae}, %zmm15, %zmm13, %zmm1
> + vmovups %zmm15, (%rsp)
> + vpmovqd %zmm9, %ymm14
> + vpmovqd %zmm8, %ymm10
> + vxorpd %zmm3, %zmm3, %zmm3{%k1}
> + vpsubd %ymm12, %ymm14, %ymm13
> + vpsubd %ymm12, %ymm10, %ymm9
> +
> +/* Polynomial. */
> + vmulpd {rn-sae}, %zmm1, %zmm1, %zmm12
> + vpcmpgtd %ymm6, %ymm13, %ymm15
> + vpcmpeqd %ymm6, %ymm13, %ymm11
> + vmulpd {rn-sae}, %zmm12, %zmm12, %zmm13
> + vpor %ymm11, %ymm15, %ymm8
> + vmovups dA19+__svml_datan2_data_internal(%rip), %zmm11
> + vmovups dA15+__svml_datan2_data_internal(%rip), %zmm15
> + vpcmpgtd %ymm6, %ymm9, %ymm14
> + vpcmpeqd %ymm6, %ymm9, %ymm6
> + vpor %ymm6, %ymm14, %ymm10
> + vmulpd {rn-sae}, %zmm13, %zmm13, %zmm14
> + vmovups dA18+__svml_datan2_data_internal(%rip), %zmm9
> + vpor %ymm10, %ymm8, %ymm6
> + vmovups dA17+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd231pd {rn-sae}, %zmm14, %zmm11, %zmm15
> + vmovups dA14+__svml_datan2_data_internal(%rip), %zmm11
> + vmovups dA12+__svml_datan2_data_internal(%rip), %zmm8
> + vfmadd231pd {rn-sae}, %zmm14, %zmm9, %zmm11
> + vmovups dA13+__svml_datan2_data_internal(%rip), %zmm9
> + vfmadd231pd {rn-sae}, %zmm14, %zmm10, %zmm9
> + vmovups dA16+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd231pd {rn-sae}, %zmm14, %zmm10, %zmm8
> + vmovups dA11+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm15
> + vmovups dA10+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm11
> + vmovups dA09+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm9
> + vmovups dA08+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm8
> + vmovups dA07+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm15
> + vmovups dA06+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm11
> + vmovups dA05+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm9
> + vmovups dA04+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm8
> + vmovups dA03+__svml_datan2_data_internal(%rip), %zmm10
> +
> +/* A00=1.0, account for it later VQFMA(D, dP4, dP4, dR8, dA00); */
> + vmulpd {rn-sae}, %zmm14, %zmm8, %zmm8
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm15
> + vmovups dA02+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm11
> + vmovups dA01+__svml_datan2_data_internal(%rip), %zmm10
> + vfmadd213pd {rn-sae}, %zmm11, %zmm12, %zmm15
> + vfmadd213pd {rn-sae}, %zmm10, %zmm14, %zmm9
> + vfmadd213pd {rn-sae}, %zmm8, %zmm12, %zmm9
> + vmovups __svml_datan2_data_internal(%rip), %zmm8
> + vfmadd213pd {rn-sae}, %zmm9, %zmm13, %zmm15
> +
> +/*
> + * Reconstruction.
> + * dP=(R+R*dP) + dPIO2
> + */
> + vfmadd213pd {rn-sae}, %zmm1, %zmm1, %zmm15
> + vaddpd {rn-sae}, %zmm3, %zmm15, %zmm1
> + vorpd %zmm5, %zmm1, %zmm9
> +
> +/* if x<0, dPI = Pi, else dPI =0 */
> + vmovups dZERO+__svml_datan2_data_internal(%rip), %zmm1
> + vcmppd $18, {sae}, %zmm1, %zmm7, %k2
> + vaddpd {rn-sae}, %zmm8, %zmm9, %zmm9{%k2}
> + vmovmskps %ymm6, %eax
> + vorpd %zmm4, %zmm9, %zmm11
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + vmovups 64(%rsp), %zmm9
> + testl %eax, %eax
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx ymm6 zmm0 zmm2 zmm3 zmm4 zmm5 zmm7 zmm9 zmm11
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm7 zmm11
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + vmovaps %zmm11, %zmm0
> + movq %rbp, %rsp
> + popq %rbp
> + cfi_def_cfa(7, 8)
> + cfi_restore(6)
> + ret
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + vmovups %zmm0, 64(%rsp)
> + vmovups %zmm7, 128(%rsp)
> + vmovups %zmm11, 192(%rsp)
> + # LOE rbx r12 r13 r14 r15 edx zmm11
> +
> + xorl %eax, %eax
> + # LOE rbx r12 r13 r14 r15 eax edx
> +
> + vzeroupper
> + movq %r12, 16(%rsp)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $8, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + vmovups 192(%rsp), %zmm11
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r12 r13 r14 r15 zmm11
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movsd 64(%rsp,%r14,8), %xmm0
> + movsd 128(%rsp,%r14,8), %xmm1
> + call atan2@PLT
> + # LOE rbx r14 r15 r12d r13d xmm0
> +
> + movsd %xmm0, 192(%rsp,%r14,8)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + vmovups dZERO+__svml_datan2_data_internal(%rip), %zmm8
> +
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + vcmppd $3, {sae}, %zmm7, %zmm7, %k1
> + vcmppd $3, {sae}, %zmm0, %zmm0, %k2
> + vcmppd $4, {sae}, %zmm8, %zmm2, %k3
> + vcmppd $4, {sae}, %zmm8, %zmm9, %k4
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + vpcmpgtq %zmm7, %zmm8, %k6
> + vpternlogd $0xff, %zmm1, %zmm1, %zmm10
> + vmovaps %zmm10, %zmm15
> + vmovaps %zmm10, %zmm12
> + vmovaps %zmm10, %zmm13
> + vpandnq %zmm2, %zmm2, %zmm15{%k3}
> + vmovaps %zmm10, %zmm2
> + vpandnq %zmm7, %zmm7, %zmm12{%k1}
> + vpandnq %zmm0, %zmm0, %zmm13{%k2}
> + vpandnq %zmm9, %zmm9, %zmm2{%k4}
> + vandpd %zmm13, %zmm12, %zmm14
> + vorpd %zmm2, %zmm15, %zmm9
> + vpsrlq $32, %zmm14, %zmm1
> + vpsrlq $32, %zmm9, %zmm2
> + vpmovqd %zmm1, %ymm1
> + vpmovqd %zmm2, %ymm9
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + vpand %ymm1, %ymm9, %ymm2
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + vmovups (%rsp), %zmm1
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + vpandn %ymm6, %ymm2, %ymm6
> + vcmppd $4, {sae}, %zmm8, %zmm1, %k5
> +
> +/* Go to callout */
> + vmovmskps %ymm6, %edx
> + vpandnq %zmm1, %zmm1, %zmm10{%k5}
> +
> +/* Set sPIO2 to zero if den. is zero */
> + vpandnq %zmm3, %zmm10, %zmm3
> + vpandq %zmm10, %zmm8, %zmm1
> + vporq %zmm1, %zmm3, %zmm3
> + vorpd %zmm5, %zmm3, %zmm1
> + vmovups __svml_datan2_data_internal(%rip), %zmm5
> + vaddpd {rn-sae}, %zmm5, %zmm1, %zmm1{%k6}
> + vorpd %zmm4, %zmm1, %zmm1
> +
> +/* Merge results from main and spec path */
> + vpmovzxdq %ymm2, %zmm4
> + vpsllq $32, %zmm4, %zmm2
> + vpord %zmm2, %zmm4, %zmm3
> + vpandnq %zmm11, %zmm3, %zmm11
> + vpandq %zmm3, %zmm1, %zmm1
> + vporq %zmm1, %zmm11, %zmm11
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm7 zmm11
> +END(_ZGVeN8vv_atan2_skx)
> +
> + .section .rodata, "a"
> + .align 64
> +
> +#ifdef __svml_datan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(64)) VUINT32 dPI[8][2];
> + __declspec(align(64)) VUINT32 dPIO2[8][2];
> + __declspec(align(64)) VUINT32 dA19[8][2];
> + __declspec(align(64)) VUINT32 dA18[8][2];
> + __declspec(align(64)) VUINT32 dA17[8][2];
> + __declspec(align(64)) VUINT32 dA16[8][2];
> + __declspec(align(64)) VUINT32 dA15[8][2];
> + __declspec(align(64)) VUINT32 dA14[8][2];
> + __declspec(align(64)) VUINT32 dA13[8][2];
> + __declspec(align(64)) VUINT32 dA12[8][2];
> + __declspec(align(64)) VUINT32 dA11[8][2];
> + __declspec(align(64)) VUINT32 dA10[8][2];
> + __declspec(align(64)) VUINT32 dA09[8][2];
> + __declspec(align(64)) VUINT32 dA08[8][2];
> + __declspec(align(64)) VUINT32 dA07[8][2];
> + __declspec(align(64)) VUINT32 dA06[8][2];
> + __declspec(align(64)) VUINT32 dA05[8][2];
> + __declspec(align(64)) VUINT32 dA04[8][2];
> + __declspec(align(64)) VUINT32 dA03[8][2];
> + __declspec(align(64)) VUINT32 dA02[8][2];
> + __declspec(align(64)) VUINT32 dA01[8][2];
> + __declspec(align(64)) VUINT32 dA00[8][2];
> + __declspec(align(64)) VUINT32 dSIGN_MASK[8][2];
> + __declspec(align(64)) VUINT32 iCHK_WORK_SUB[16][1];
> + __declspec(align(64)) VUINT32 iCHK_WORK_CMP[16][1];
> + __declspec(align(64)) VUINT32 dABS_MASK[8][2];
> + __declspec(align(64)) VUINT32 dZERO[8][2];
> +} __svml_datan2_data_internal;
> +#endif
> +__svml_datan2_data_internal:
> + .quad 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18, 0x400921FB54442D18 //dPI
> + .align 64
> + .quad 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18, 0x3FF921FB54442D18 //dPIO2
> + .align 64
> + .quad 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3 // dA19
> + .align 64
> + .quad 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209, 0x3F2CED0A36665209 // dA18
> + .align 64
> + .quad 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23, 0xBF52E67C93954C23 // dA17
> + .align 64
> + .quad 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3 // dA16
> + .align 64
> + .quad 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD // dA15
> + .align 64
> + .quad 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5, 0x3F914F4C661116A5 // dA14
> + .align 64
> + .quad 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C // dA13
> + .align 64
> + .quad 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F // dA12
> + .align 64
> + .quad 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC // dA11
> + .align 64
> + .quad 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B // dA10
> + .align 64
> + .quad 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954, 0xBFAAD261EAA09954 // dA09
> + .align 64
> + .quad 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF // dA08
> + .align 64
> + .quad 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0, 0xBFB11084009435E0 // dA07
> + .align 64
> + .quad 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651, 0x3FB3B12A49295651 // dA06
> + .align 64
> + .quad 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94, 0xBFB745D009BADA94 // dA05
> + .align 64
> + .quad 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5 // dA04
> + .align 64
> + .quad 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7, 0xBFC2492491EE55C7 // dA03
> + .align 64
> + .quad 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34, 0x3FC999999997EE34 // dA02
> + .align 64
> + .quad 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5, 0xBFD55555555553C5 // dA01
> + .align 64
> + .quad 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000 // dA00
> + .align 64
> + .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 //dSIGN_MASK
> + .align 64
> + .long 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000, 0x80300000 //iCHK_WORK_SUB
> + .align 64
> + .long 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000 //iCHK_WORK_CMP
> + .align 64
> + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff //dABS_MASK
> + .align 64
> + .quad 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000 //dZERO
> + .align 64
> + .type __svml_datan2_data_internal,@object
> + .size __svml_datan2_data_internal,.-__svml_datan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core-avx2.S
> new file mode 100644
> index 0000000000..a2a76e8bfd
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core-avx2.S
> @@ -0,0 +1,20 @@
> +/* AVX2 version of vectorized atan2f.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVeN16vv_atan2f _ZGVeN16vv_atan2f_avx2_wrapper
> +#include "../svml_s_atan2f16_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core.c
> new file mode 100644
> index 0000000000..6fa806414d
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized atan2f, vector length is 16.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVeN16vv_atan2f
> +#include "ifunc-mathvec-avx512-skx.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVeN16vv_atan2f, __GI__ZGVeN16vv_atan2f,
> + __redirect__ZGVeN16vv_atan2f)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core_avx512.S
> new file mode 100644
> index 0000000000..f3477cc8e6
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f16_core_avx512.S
> @@ -0,0 +1,399 @@
> +/* Function atan2f vectorized with AVX-512.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_satan2_data_internal
> + */
> +#define sZERO 0
> +#define sONE 64
> +#define sSIGN_MASK 128
> +#define sABS_MASK 192
> +#define sPIO2 256
> +#define sPI 320
> +#define sPC8 384
> +#define sPC7 448
> +#define sPC6 512
> +#define sPC5 576
> +#define sPC4 640
> +#define sPC3 704
> +#define sPC2 768
> +#define sPC1 832
> +#define sPC0 896
> +#define iCHK_WORK_SUB 960
> +#define iCHK_WORK_CMP 1024
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.exex512,"ax",@progbits
> +ENTRY(_ZGVeN16vv_atan2f_skx)
> + pushq %rbp
> + cfi_def_cfa_offset(16)
> + movq %rsp, %rbp
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> + andq $-64, %rsp
> + subq $256, %rsp
> + xorl %edx, %edx
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Arguments signs
> + */
> + vmovups sABS_MASK+__svml_satan2_data_internal(%rip), %zmm6
> + vmovups sONE+__svml_satan2_data_internal(%rip), %zmm3
> +
> +/* Testing on working interval. */
> + vmovups iCHK_WORK_SUB+__svml_satan2_data_internal(%rip), %zmm9
> + vmovups iCHK_WORK_CMP+__svml_satan2_data_internal(%rip), %zmm14
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + vmovups sPIO2+__svml_satan2_data_internal(%rip), %zmm4
> + vpternlogd $255, %zmm13, %zmm13, %zmm13
> + vmovaps %zmm1, %zmm8
> + vandps %zmm6, %zmm8, %zmm2
> + vandps %zmm6, %zmm0, %zmm1
> + vorps sSIGN_MASK+__svml_satan2_data_internal(%rip), %zmm2, %zmm5
> + vpsubd %zmm9, %zmm2, %zmm10
> + vpsubd %zmm9, %zmm1, %zmm12
> + vxorps %zmm2, %zmm8, %zmm7
> + vxorps %zmm1, %zmm0, %zmm6
> + vcmpps $17, {sae}, %zmm2, %zmm1, %k1
> + vpcmpgtd %zmm10, %zmm14, %k2
> + vpcmpgtd %zmm12, %zmm14, %k3
> + vmovups sPC6+__svml_satan2_data_internal(%rip), %zmm14
> + vblendmps %zmm1, %zmm5, %zmm11{%k1}
> + vblendmps %zmm2, %zmm1, %zmm5{%k1}
> + vxorps %zmm4, %zmm4, %zmm4{%k1}
> +
> +/*
> + * Division a/b.
> + * Enabled when FMA is available and
> + * performance is better with NR iteration
> + */
> + vrcp14ps %zmm5, %zmm15
> + vfnmadd231ps {rn-sae}, %zmm5, %zmm15, %zmm3
> + vfmadd213ps {rn-sae}, %zmm15, %zmm3, %zmm15
> + vmulps {rn-sae}, %zmm15, %zmm11, %zmm3
> + vfnmadd231ps {rn-sae}, %zmm5, %zmm3, %zmm11
> + vfmadd213ps {rn-sae}, %zmm3, %zmm11, %zmm15
> + vmovups sPC8+__svml_satan2_data_internal(%rip), %zmm11
> + vpternlogd $255, %zmm3, %zmm3, %zmm3
> +
> +/* Polynomial. */
> + vmulps {rn-sae}, %zmm15, %zmm15, %zmm9
> + vpandnd %zmm10, %zmm10, %zmm13{%k2}
> + vmulps {rn-sae}, %zmm9, %zmm9, %zmm10
> + vfmadd231ps {rn-sae}, %zmm10, %zmm11, %zmm14
> + vmovups sPC5+__svml_satan2_data_internal(%rip), %zmm11
> + vpandnd %zmm12, %zmm12, %zmm3{%k3}
> + vpord %zmm3, %zmm13, %zmm3
> + vmovups sPC4+__svml_satan2_data_internal(%rip), %zmm13
> + vmovups sPC7+__svml_satan2_data_internal(%rip), %zmm12
> + vptestmd %zmm3, %zmm3, %k0
> + vfmadd213ps {rn-sae}, %zmm13, %zmm10, %zmm14
> + vfmadd231ps {rn-sae}, %zmm10, %zmm12, %zmm11
> + vmovups sPC3+__svml_satan2_data_internal(%rip), %zmm12
> + vmovups sPC2+__svml_satan2_data_internal(%rip), %zmm13
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + kortestw %k0, %k0
> + vfmadd213ps {rn-sae}, %zmm12, %zmm10, %zmm11
> + vmovups sPC1+__svml_satan2_data_internal(%rip), %zmm12
> + vfmadd213ps {rn-sae}, %zmm13, %zmm10, %zmm14
> + vmovups sPC0+__svml_satan2_data_internal(%rip), %zmm13
> + vfmadd213ps {rn-sae}, %zmm12, %zmm10, %zmm11
> + vfmadd213ps {rn-sae}, %zmm13, %zmm10, %zmm14
> + vfmadd213ps {rn-sae}, %zmm14, %zmm9, %zmm11
> +
> +/* Reconstruction. */
> + vfmadd213ps {rn-sae}, %zmm4, %zmm15, %zmm11
> +
> +/* if x<0, sPI = Pi, else sPI =0 */
> + vmovups __svml_satan2_data_internal(%rip), %zmm15
> + vorps %zmm7, %zmm11, %zmm9
> + vcmpps $18, {sae}, %zmm15, %zmm8, %k4
> + vmovups sPI+__svml_satan2_data_internal(%rip), %zmm11
> + vaddps {rn-sae}, %zmm11, %zmm9, %zmm9{%k4}
> + vorps %zmm6, %zmm9, %zmm10
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm1 zmm2 zmm3 zmm4 zmm5 zmm6 zmm7 zmm8 zmm10 zmm11
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm8 zmm10
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + vmovaps %zmm10, %zmm0
> + movq %rbp, %rsp
> + popq %rbp
> + cfi_def_cfa(7, 8)
> + cfi_restore(6)
> + ret
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + vmovups %zmm0, 64(%rsp)
> + vmovups %zmm8, 128(%rsp)
> + vmovups %zmm10, 192(%rsp)
> + # LOE rbx r12 r13 r14 r15 edx zmm10
> +
> + xorl %eax, %eax
> + # LOE rbx r12 r13 r14 r15 eax edx
> +
> + vzeroupper
> + movq %r12, 16(%rsp)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $16, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + vmovups 192(%rsp), %zmm10
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r12 r13 r14 r15 zmm10
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movss 64(%rsp,%r14,4), %xmm0
> + movss 128(%rsp,%r14,4), %xmm1
> + call atan2f@PLT
> + # LOE rbx r14 r15 r12d r13d xmm0
> +
> + movss %xmm0, 192(%rsp,%r14,4)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + vmovups __svml_satan2_data_internal(%rip), %zmm9
> +
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + vcmpps $3, {sae}, %zmm8, %zmm8, %k1
> + vcmpps $3, {sae}, %zmm0, %zmm0, %k2
> + vpcmpd $4, %zmm9, %zmm2, %k3
> + vpcmpd $4, %zmm9, %zmm1, %k4
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + vcmpps $4, {sae}, %zmm9, %zmm5, %k5
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + vpcmpgtd %zmm8, %zmm9, %k6
> + vpternlogd $255, %zmm14, %zmm14, %zmm14
> + vpternlogd $255, %zmm12, %zmm12, %zmm12
> + vpternlogd $255, %zmm13, %zmm13, %zmm13
> + vpandnd %zmm2, %zmm2, %zmm14{%k3}
> + vpternlogd $255, %zmm2, %zmm2, %zmm2
> + vpandnd %zmm1, %zmm1, %zmm2{%k4}
> + vpord %zmm2, %zmm14, %zmm15
> + vpternlogd $255, %zmm2, %zmm2, %zmm2
> + vpandnd %zmm5, %zmm5, %zmm2{%k5}
> +
> +/* Set sPIO2 to zero if den. is zero */
> + vpandnd %zmm4, %zmm2, %zmm4
> + vpandd %zmm2, %zmm9, %zmm5
> + vpord %zmm5, %zmm4, %zmm2
> + vorps %zmm7, %zmm2, %zmm7
> + vaddps {rn-sae}, %zmm11, %zmm7, %zmm7{%k6}
> + vorps %zmm6, %zmm7, %zmm6
> + vpandnd %zmm8, %zmm8, %zmm12{%k1}
> + vpandnd %zmm0, %zmm0, %zmm13{%k2}
> + vandps %zmm13, %zmm12, %zmm12
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + vpandd %zmm12, %zmm15, %zmm1
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + vpandnd %zmm3, %zmm1, %zmm3
> +
> +/* Go to callout */
> + vptestmd %zmm3, %zmm3, %k0
> + kmovw %k0, %edx
> +
> +/* Merge results from main and spec path */
> + vpandnd %zmm10, %zmm1, %zmm10
> + vpandd %zmm1, %zmm6, %zmm11
> + vpord %zmm11, %zmm10, %zmm10
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm8 zmm10
> +END(_ZGVeN16vv_atan2f_skx)
> +
> + .section .rodata, "a"
> + .align 64
> +
> +#ifdef __svml_satan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(64)) VUINT32 sZERO[16][1];
> + __declspec(align(64)) VUINT32 sONE[16][1];
> + __declspec(align(64)) VUINT32 sSIGN_MASK[16][1];
> + __declspec(align(64)) VUINT32 sABS_MASK[16][1];
> + __declspec(align(64)) VUINT32 sPIO2[16][1];
> + __declspec(align(64)) VUINT32 sPI[16][1];
> + __declspec(align(64)) VUINT32 sPC8[16][1];
> + __declspec(align(64)) VUINT32 sPC7[16][1];
> + __declspec(align(64)) VUINT32 sPC6[16][1];
> + __declspec(align(64)) VUINT32 sPC5[16][1];
> + __declspec(align(64)) VUINT32 sPC4[16][1];
> + __declspec(align(64)) VUINT32 sPC3[16][1];
> + __declspec(align(64)) VUINT32 sPC2[16][1];
> + __declspec(align(64)) VUINT32 sPC1[16][1];
> + __declspec(align(64)) VUINT32 sPC0[16][1];
> + __declspec(align(64)) VUINT32 iCHK_WORK_SUB[16][1];
> + __declspec(align(64)) VUINT32 iCHK_WORK_CMP[16][1];
> +} __svml_satan2_data_internal;
> +#endif
> +__svml_satan2_data_internal:
> + .long 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000 // sZERO
> + .align 64
> + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 // sONE
> + .align 64
> + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 // sSIGN_MASK
> + .align 64
> + .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF // sABS_MASK
> + .align 64
> + .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB // sPIO2
> + .align 64
> + .long 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB // sPI
> + .align 64
> + .long 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0 // sA08
> + .align 64
> + .long 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631 // sA07
> + .align 64
> + .long 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384 // sA06
> + .align 64
> + .long 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629 // sA05
> + .align 64
> + .long 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474 // sA04
> + .align 64
> + .long 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8 // sA03
> + .align 64
> + .long 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F // sA02
> + .align 64
> + .long 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49 // sA01
> + .align 64
> + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 // sA00
> + .align 64
> + .long 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000 //iCHK_WORK_SUB
> + .align 64
> + .long 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000 //iCHK_WORK_CMP
> + .align 64
> + .type __svml_satan2_data_internal,@object
> + .size __svml_satan2_data_internal,.-__svml_satan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core-sse2.S
> new file mode 100644
> index 0000000000..d1a67facf1
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core-sse2.S
> @@ -0,0 +1,20 @@
> +/* SSE2 version of vectorized atan2f.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVbN4vv_atan2f _ZGVbN4vv_atan2f_sse2
> +#include "../svml_s_atan2f4_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core.c
> new file mode 100644
> index 0000000000..ee882b0557
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized atan2f, vector length is 4.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVbN4vv_atan2f
> +#include "ifunc-mathvec-sse4_1.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVbN4vv_atan2f, __GI__ZGVbN4vv_atan2f,
> + __redirect__ZGVbN4vv_atan2f)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core_sse4.S
> new file mode 100644
> index 0000000000..e4fbe82501
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f4_core_sse4.S
> @@ -0,0 +1,384 @@
> +/* Function atan2f vectorized with SSE4.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_satan2_data_internal
> + */
> +#define sZERO 0
> +#define sSIGN_MASK 16
> +#define sABS_MASK 32
> +#define sPIO2 48
> +#define sPI 64
> +#define sPC8 80
> +#define sPC7 96
> +#define sPC6 112
> +#define sPC5 128
> +#define sPC4 144
> +#define sPC3 160
> +#define sPC2 176
> +#define sPC1 192
> +#define sPC0 208
> +#define iCHK_WORK_SUB 224
> +#define iCHK_WORK_CMP 240
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.sse4,"ax",@progbits
> +ENTRY(_ZGVbN4vv_atan2f_sse4)
> + subq $88, %rsp
> + cfi_def_cfa_offset(96)
> + movaps %xmm0, %xmm12
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Arguments signs
> + */
> + movups sABS_MASK+__svml_satan2_data_internal(%rip), %xmm10
> + movaps %xmm1, %xmm13
> + movaps %xmm10, %xmm11
> + andps %xmm12, %xmm10
> + andps %xmm13, %xmm11
> + movaps %xmm10, %xmm7
> + cmpltps %xmm11, %xmm7
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + movups sSIGN_MASK+__svml_satan2_data_internal(%rip), %xmm6
> + movaps %xmm7, %xmm0
> + orps %xmm11, %xmm6
> + movaps %xmm10, %xmm4
> + andnps %xmm6, %xmm0
> + movaps %xmm7, %xmm6
> + movaps %xmm11, %xmm5
> + andps %xmm7, %xmm4
> + andnps %xmm10, %xmm6
> + andps %xmm7, %xmm5
> + orps %xmm4, %xmm0
> + orps %xmm5, %xmm6
> +
> +/* Division a/b. */
> + divps %xmm6, %xmm0
> +
> +/* Testing on working interval. */
> + movdqu iCHK_WORK_SUB+__svml_satan2_data_internal(%rip), %xmm14
> + movaps %xmm11, %xmm15
> + movaps %xmm10, %xmm3
> + psubd %xmm14, %xmm15
> + psubd %xmm14, %xmm3
> + movdqa %xmm15, %xmm1
> + movdqu iCHK_WORK_CMP+__svml_satan2_data_internal(%rip), %xmm2
> + movdqa %xmm3, %xmm14
> + pcmpgtd %xmm2, %xmm1
> + pcmpeqd %xmm2, %xmm15
> + pcmpgtd %xmm2, %xmm14
> + pcmpeqd %xmm2, %xmm3
> + por %xmm15, %xmm1
> + por %xmm3, %xmm14
> + por %xmm14, %xmm1
> +
> +/* Polynomial. */
> + movaps %xmm0, %xmm14
> + mulps %xmm0, %xmm14
> + movaps %xmm13, %xmm4
> + movmskps %xmm1, %ecx
> + movaps %xmm14, %xmm15
> + movaps %xmm11, %xmm9
> + mulps %xmm14, %xmm15
> + pxor %xmm13, %xmm9
> + movups sPC8+__svml_satan2_data_internal(%rip), %xmm2
> + movaps %xmm10, %xmm8
> + mulps %xmm15, %xmm2
> + pxor %xmm12, %xmm8
> + movups sPC7+__svml_satan2_data_internal(%rip), %xmm3
> + xorl %edx, %edx
> + mulps %xmm15, %xmm3
> + addps sPC6+__svml_satan2_data_internal(%rip), %xmm2
> + mulps %xmm15, %xmm2
> + addps sPC5+__svml_satan2_data_internal(%rip), %xmm3
> + mulps %xmm15, %xmm3
> + addps sPC4+__svml_satan2_data_internal(%rip), %xmm2
> + mulps %xmm15, %xmm2
> + addps sPC3+__svml_satan2_data_internal(%rip), %xmm3
> + mulps %xmm15, %xmm3
> + addps sPC2+__svml_satan2_data_internal(%rip), %xmm2
> + mulps %xmm2, %xmm15
> + addps sPC1+__svml_satan2_data_internal(%rip), %xmm3
> + mulps %xmm3, %xmm14
> + addps sPC0+__svml_satan2_data_internal(%rip), %xmm15
> +
> +/* if x<0, sPI = Pi, else sPI =0 */
> + movups __svml_satan2_data_internal(%rip), %xmm5
> + xorl %eax, %eax
> + andnps sPIO2+__svml_satan2_data_internal(%rip), %xmm7
> + addps %xmm14, %xmm15
> + cmpleps %xmm5, %xmm4
> +
> +/* Reconstruction. */
> + mulps %xmm15, %xmm0
> + andps sPI+__svml_satan2_data_internal(%rip), %xmm4
> + addps %xmm7, %xmm0
> + orps %xmm9, %xmm0
> + addps %xmm4, %xmm0
> + orps %xmm8, %xmm0
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + testl %ecx, %ecx
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm1 xmm4 xmm5 xmm6 xmm7 xmm8 xmm9 xmm10 xmm11 xmm12 xmm13
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm12 xmm13
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + addq $88, %rsp
> + cfi_def_cfa_offset(8)
> + ret
> + cfi_def_cfa_offset(96)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + movups %xmm12, 32(%rsp)
> + movups %xmm13, 48(%rsp)
> + movups %xmm0, 64(%rsp)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0
> +
> + movq %r12, 16(%rsp)
> + cfi_offset(12, -80)
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + cfi_offset(13, -88)
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + cfi_offset(14, -96)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $4, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx rbp r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + movups 64(%rsp), %xmm0
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + cfi_offset(12, -80)
> + cfi_offset(13, -88)
> + cfi_offset(14, -96)
> + # LOE rbx rbp r12 r13 r14 r15 xmm0
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movss 32(%rsp,%r14,4), %xmm0
> + movss 48(%rsp,%r14,4), %xmm1
> + call atan2f@PLT
> + # LOE rbx rbp r14 r15 r12d r13d xmm0
> +
> + movss %xmm0, 64(%rsp,%r14,4)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx rbp r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + movaps %xmm13, %xmm3
> + movaps %xmm12, %xmm2
> + cmpordps %xmm13, %xmm3
> + cmpordps %xmm12, %xmm2
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + cmpeqps %xmm5, %xmm6
> +
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + pcmpeqd %xmm5, %xmm11
> + pcmpeqd %xmm5, %xmm10
> + andps %xmm2, %xmm3
> + por %xmm10, %xmm11
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + andps %xmm3, %xmm11
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + movaps %xmm11, %xmm10
> + pandn %xmm1, %xmm10
> +
> +/* Set sPIO2 to zero if den. is zero */
> + movaps %xmm6, %xmm1
> + andnps %xmm7, %xmm1
> + andps %xmm5, %xmm6
> + orps %xmm6, %xmm1
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + pcmpgtd %xmm13, %xmm5
> + orps %xmm9, %xmm1
> + andps %xmm4, %xmm5
> +
> +/* Merge results from main and spec path */
> + movaps %xmm11, %xmm4
> + addps %xmm5, %xmm1
> +
> +/* Go to callout */
> + movmskps %xmm10, %edx
> + orps %xmm8, %xmm1
> + andnps %xmm0, %xmm4
> + andps %xmm11, %xmm1
> + movaps %xmm4, %xmm0
> + orps %xmm1, %xmm0
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm12 xmm13
> +END(_ZGVbN4vv_atan2f_sse4)
> +
> + .section .rodata, "a"
> + .align 16
> +
> +#ifdef __svml_satan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(16)) VUINT32 sZERO[4][1];
> + __declspec(align(16)) VUINT32 sSIGN_MASK[4][1];
> + __declspec(align(16)) VUINT32 sABS_MASK[4][1];
> + __declspec(align(16)) VUINT32 sPIO2[4][1];
> + __declspec(align(16)) VUINT32 sPI[4][1];
> + __declspec(align(16)) VUINT32 sPC8[4][1];
> + __declspec(align(16)) VUINT32 sPC7[4][1];
> + __declspec(align(16)) VUINT32 sPC6[4][1];
> + __declspec(align(16)) VUINT32 sPC5[4][1];
> + __declspec(align(16)) VUINT32 sPC4[4][1];
> + __declspec(align(16)) VUINT32 sPC3[4][1];
> + __declspec(align(16)) VUINT32 sPC2[4][1];
> + __declspec(align(16)) VUINT32 sPC1[4][1];
> + __declspec(align(16)) VUINT32 sPC0[4][1];
> + __declspec(align(16)) VUINT32 iCHK_WORK_SUB[4][1];
> + __declspec(align(16)) VUINT32 iCHK_WORK_CMP[4][1];
> +} __svml_satan2_data_internal;
> +#endif
> +__svml_satan2_data_internal:
> + .long 0x00000000, 0x00000000, 0x00000000, 0x00000000 // sZERO
> + .align 16
> + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000 // sSIGN_MASK
> + .align 16
> + .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF // sABS_MASK
> + .align 16
> + .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB // sPIO2
> + .align 16
> + .long 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB // sPI
> + .align 16
> + .long 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0 // sA08
> + .align 16
> + .long 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631 // sA07
> + .align 16
> + .long 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384 // sA06
> + .align 16
> + .long 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629 // sA05
> + .align 16
> + .long 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474 // sA04
> + .align 16
> + .long 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8 // sA03
> + .align 16
> + .long 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F // sA02
> + .align 16
> + .long 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49 // sA01
> + .align 16
> + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 // sA00
> + .align 16
> + .long 0x81000000, 0x81000000, 0x81000000, 0x81000000 //iCHK_WORK_SUB
> + .align 16
> + .long 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000 //iCHK_WORK_CMP
> + .align 16
> + .type __svml_satan2_data_internal,@object
> + .size __svml_satan2_data_internal,.-__svml_satan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core-sse.S
> new file mode 100644
> index 0000000000..21b1d3ff63
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core-sse.S
> @@ -0,0 +1,20 @@
> +/* SSE version of vectorized atan2f.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define _ZGVdN8vv_atan2f _ZGVdN8vv_atan2f_sse_wrapper
> +#include "../svml_s_atan2f8_core.S"
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core.c
> new file mode 100644
> index 0000000000..7e02050983
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core.c
> @@ -0,0 +1,28 @@
> +/* Multiple versions of vectorized sinf, vector length is 8.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#define SYMBOL_NAME _ZGVdN8vv_atan2f
> +#include "ifunc-mathvec-avx2.h"
> +
> +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ());
> +
> +#ifdef SHARED
> +__hidden_ver1 (_ZGVdN8vv_atan2f, __GI__ZGVdN8vv_atan2f,
> + __redirect__ZGVdN8vv_atan2f)
> + __attribute__ ((visibility ("hidden")));
> +#endif
> diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core_avx2.S
> new file mode 100644
> index 0000000000..2e6e5eb71c
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atan2f8_core_avx2.S
> @@ -0,0 +1,362 @@
> +/* Function atan2f vectorized with AVX2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + https://www.gnu.org/licenses/. */
> +
> +/*
> + * ALGORITHM DESCRIPTION:
> + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
> + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
> + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
> + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
> + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
> + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
> + *
> + *
> + */
> +
> +/* Offsets for data table __svml_satan2_data_internal
> + */
> +#define sZERO 0
> +#define sSIGN_MASK 32
> +#define sABS_MASK 64
> +#define sPIO2 96
> +#define sPI 128
> +#define sPC8 160
> +#define sPC7 192
> +#define sPC6 224
> +#define sPC5 256
> +#define sPC4 288
> +#define sPC3 320
> +#define sPC2 352
> +#define sPC1 384
> +#define sPC0 416
> +#define iCHK_WORK_SUB 448
> +#define iCHK_WORK_CMP 480
> +
> +#include <sysdep.h>
> +
> + .text
> + .section .text.avx2,"ax",@progbits
> +ENTRY(_ZGVdN8vv_atan2f_avx2)
> + pushq %rbp
> + cfi_def_cfa_offset(16)
> + movq %rsp, %rbp
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> + andq $-32, %rsp
> + subq $128, %rsp
> + xorl %edx, %edx
> +
> +/*
> + * #define NO_VECTOR_ZERO_ATAN2_ARGS
> + * Declarations
> + * Variables
> + * Constants
> + * The end of declarations
> + * Implementation
> + * Arguments signs
> + */
> + vmovups sABS_MASK+__svml_satan2_data_internal(%rip), %ymm2
> +
> +/* Testing on working interval. */
> + vmovups iCHK_WORK_SUB+__svml_satan2_data_internal(%rip), %ymm15
> + vmovups iCHK_WORK_CMP+__svml_satan2_data_internal(%rip), %ymm9
> +
> +/* if x<0, sPI = Pi, else sPI =0 */
> + vmovups __svml_satan2_data_internal(%rip), %ymm5
> + vmovaps %ymm1, %ymm7
> + vandps %ymm2, %ymm7, %ymm13
> + vandps %ymm2, %ymm0, %ymm12
> + vcmplt_oqps %ymm13, %ymm12, %ymm4
> + vcmple_oqps %ymm5, %ymm7, %ymm6
> + vpsubd %ymm15, %ymm13, %ymm10
> + vpsubd %ymm15, %ymm12, %ymm8
> +
> +/*
> + * 1) If y<x then a= y, b=x, PIO2=0
> + * 2) If y>x then a=-x, b=y, PIO2=Pi/2
> + */
> + vorps sSIGN_MASK+__svml_satan2_data_internal(%rip), %ymm13, %ymm3
> + vblendvps %ymm4, %ymm12, %ymm3, %ymm14
> + vblendvps %ymm4, %ymm13, %ymm12, %ymm3
> +
> +/* Division a/b. */
> + vdivps %ymm3, %ymm14, %ymm11
> + vpcmpgtd %ymm9, %ymm10, %ymm14
> + vpcmpeqd %ymm9, %ymm10, %ymm15
> + vpor %ymm15, %ymm14, %ymm10
> + vmovups sPC7+__svml_satan2_data_internal(%rip), %ymm15
> + vpcmpgtd %ymm9, %ymm8, %ymm14
> + vpcmpeqd %ymm9, %ymm8, %ymm8
> + vpor %ymm8, %ymm14, %ymm9
> + vmovups sPC8+__svml_satan2_data_internal(%rip), %ymm14
> + vpor %ymm9, %ymm10, %ymm10
> +
> +/* Polynomial. */
> + vmulps %ymm11, %ymm11, %ymm9
> + vmulps %ymm9, %ymm9, %ymm8
> + vfmadd213ps sPC6+__svml_satan2_data_internal(%rip), %ymm8, %ymm14
> + vfmadd213ps sPC5+__svml_satan2_data_internal(%rip), %ymm8, %ymm15
> + vfmadd213ps sPC4+__svml_satan2_data_internal(%rip), %ymm8, %ymm14
> + vfmadd213ps sPC3+__svml_satan2_data_internal(%rip), %ymm8, %ymm15
> + vfmadd213ps sPC2+__svml_satan2_data_internal(%rip), %ymm8, %ymm14
> + vfmadd213ps sPC1+__svml_satan2_data_internal(%rip), %ymm8, %ymm15
> + vfmadd213ps sPC0+__svml_satan2_data_internal(%rip), %ymm8, %ymm14
> + vfmadd213ps %ymm14, %ymm9, %ymm15
> + vandnps sPIO2+__svml_satan2_data_internal(%rip), %ymm4, %ymm4
> +
> +/* Reconstruction. */
> + vfmadd213ps %ymm4, %ymm11, %ymm15
> + vxorps %ymm13, %ymm7, %ymm1
> + vandps sPI+__svml_satan2_data_internal(%rip), %ymm6, %ymm6
> + vorps %ymm1, %ymm15, %ymm11
> + vaddps %ymm11, %ymm6, %ymm8
> + vmovmskps %ymm10, %eax
> + vxorps %ymm12, %ymm0, %ymm2
> + vorps %ymm2, %ymm8, %ymm9
> +
> +/* Special branch for fast (vector) processing of zero arguments */
> + testl %eax, %eax
> +
> +/* Go to auxilary branch */
> + jne L(AUX_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm1 ymm2 ymm3 ymm4 ymm5 ymm6 ymm7 ymm9 ymm10 ymm12 ymm13
> +
> +/* Return from auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH_RETURN):
> +/*
> + * Special branch for fast (vector) processing of zero arguments
> + * The end of implementation
> + */
> + testl %edx, %edx
> +
> +/* Go to special inputs processing branch */
> + jne L(SPECIAL_VALUES_BRANCH)
> + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm7 ymm9
> +
> +/* Restore registers
> + * and exit the function
> + */
> +
> +L(EXIT):
> + vmovaps %ymm9, %ymm0
> + movq %rbp, %rsp
> + popq %rbp
> + cfi_def_cfa(7, 8)
> + cfi_restore(6)
> + ret
> + cfi_def_cfa(6, 16)
> + cfi_offset(6, -16)
> +
> +/* Branch to process
> + * special inputs
> + */
> +
> +L(SPECIAL_VALUES_BRANCH):
> + vmovups %ymm0, 32(%rsp)
> + vmovups %ymm7, 64(%rsp)
> + vmovups %ymm9, 96(%rsp)
> + # LOE rbx r12 r13 r14 r15 edx ymm9
> +
> + xorl %eax, %eax
> + # LOE rbx r12 r13 r14 r15 eax edx
> +
> + vzeroupper
> + movq %r12, 16(%rsp)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22
> + movl %eax, %r12d
> + movq %r13, 8(%rsp)
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
> + movl %edx, %r13d
> + movq %r14, (%rsp)
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r15 r12d r13d
> +
> +/* Range mask
> + * bits check
> + */
> +
> +L(RANGEMASK_CHECK):
> + btl %r12d, %r13d
> +
> +/* Call scalar math function */
> + jc L(SCALAR_MATH_CALL)
> + # LOE rbx r15 r12d r13d
> +
> +/* Special inputs
> + * processing loop
> + */
> +
> +L(SPECIAL_VALUES_LOOP):
> + incl %r12d
> + cmpl $8, %r12d
> +
> +/* Check bits in range mask */
> + jl L(RANGEMASK_CHECK)
> + # LOE rbx r15 r12d r13d
> +
> + movq 16(%rsp), %r12
> + cfi_restore(12)
> + movq 8(%rsp), %r13
> + cfi_restore(13)
> + movq (%rsp), %r14
> + cfi_restore(14)
> + vmovups 96(%rsp), %ymm9
> +
> +/* Go to exit */
> + jmp L(EXIT)
> + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
> + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */
> + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
> + # LOE rbx r12 r13 r14 r15 ymm9
> +
> +/* Scalar math fucntion call
> + * to process special input
> + */
> +
> +L(SCALAR_MATH_CALL):
> + movl %r12d, %r14d
> + movss 32(%rsp,%r14,4), %xmm0
> + movss 64(%rsp,%r14,4), %xmm1
> + call atan2f@PLT
> + # LOE rbx r14 r15 r12d r13d xmm0
> +
> + movss %xmm0, 96(%rsp,%r14,4)
> +
> +/* Process special inputs in loop */
> + jmp L(SPECIAL_VALUES_LOOP)
> + cfi_restore(12)
> + cfi_restore(13)
> + cfi_restore(14)
> + # LOE rbx r15 r12d r13d
> +
> +/* Auxilary branch
> + * for out of main path inputs
> + */
> +
> +L(AUX_BRANCH):
> +/* Check if at least on of Y or Y is zero: iAXAYZERO */
> + vpcmpeqd %ymm5, %ymm13, %ymm13
> + vpcmpeqd %ymm5, %ymm12, %ymm12
> +
> +/* Check if both X & Y are not NaNs: iXYnotNAN */
> + vcmpordps %ymm7, %ymm7, %ymm11
> + vcmpordps %ymm0, %ymm0, %ymm14
> +
> +/*
> + * Path for zero arguments (at least one of both)
> + * Check if both args are zeros (den. is zero)
> + */
> + vcmpeqps %ymm5, %ymm3, %ymm3
> + vpor %ymm12, %ymm13, %ymm15
> +
> +/* Set sPIO2 to zero if den. is zero */
> + vblendvps %ymm3, %ymm5, %ymm4, %ymm4
> + vandps %ymm14, %ymm11, %ymm8
> +
> +/* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
> + vpand %ymm8, %ymm15, %ymm8
> +
> +/* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
> + vpcmpgtd %ymm7, %ymm5, %ymm5
> + vorps %ymm1, %ymm4, %ymm1
> + vandps %ymm6, %ymm5, %ymm6
> + vaddps %ymm6, %ymm1, %ymm1
> +
> +/* Exclude from previous callout mask zero (and not NaN) arguments */
> + vpandn %ymm10, %ymm8, %ymm10
> + vorps %ymm2, %ymm1, %ymm2
> +
> +/* Go to callout */
> + vmovmskps %ymm10, %edx
> +
> +/* Merge results from main and spec path */
> + vblendvps %ymm8, %ymm2, %ymm9, %ymm9
> +
> +/* Return to main vector processing path */
> + jmp L(AUX_BRANCH_RETURN)
> + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm7 ymm9
> +END(_ZGVdN8vv_atan2f_avx2)
> +
> + .section .rodata, "a"
> + .align 32
> +
> +#ifdef __svml_satan2_data_internal_typedef
> +typedef unsigned int VUINT32;
> +typedef struct {
> + __declspec(align(32)) VUINT32 sZERO[8][1];
> + __declspec(align(32)) VUINT32 sSIGN_MASK[8][1];
> + __declspec(align(32)) VUINT32 sABS_MASK[8][1];
> + __declspec(align(32)) VUINT32 sPIO2[8][1];
> + __declspec(align(32)) VUINT32 sPI[8][1];
> + __declspec(align(32)) VUINT32 sPC8[8][1];
> + __declspec(align(32)) VUINT32 sPC7[8][1];
> + __declspec(align(32)) VUINT32 sPC6[8][1];
> + __declspec(align(32)) VUINT32 sPC5[8][1];
> + __declspec(align(32)) VUINT32 sPC4[8][1];
> + __declspec(align(32)) VUINT32 sPC3[8][1];
> + __declspec(align(32)) VUINT32 sPC2[8][1];
> + __declspec(align(32)) VUINT32 sPC1[8][1];
> + __declspec(align(32)) VUINT32 sPC0[8][1];
> + __declspec(align(32)) VUINT32 iCHK_WORK_SUB[8][1];
> + __declspec(align(32)) VUINT32 iCHK_WORK_CMP[8][1];
> +} __svml_satan2_data_internal;
> +#endif
> +__svml_satan2_data_internal:
> + .long 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000 // sZERO
> + .align 32
> + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 // sSIGN_MASK
> + .align 32
> + .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF // sABS_MASK
> + .align 32
> + .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB // sPIO2
> + .align 32
> + .long 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB, 0x40490FDB // sPI
> + .align 32
> + .long 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0 // sA08
> + .align 32
> + .long 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631 // sA07
> + .align 32
> + .long 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384 // sA06
> + .align 32
> + .long 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629 // sA05
> + .align 32
> + .long 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474 // sA04
> + .align 32
> + .long 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8 // sA03
> + .align 32
> + .long 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F // sA02
> + .align 32
> + .long 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49 // sA01
> + .align 32
> + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 // sA00
> + .align 32
> + .long 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000, 0x81000000 //iCHK_WORK_SUB
> + .align 32
> + .long 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000, 0xFC000000 //iCHK_WORK_CMP
> + .align 32
> + .type __svml_satan2_data_internal,@object
> + .size __svml_satan2_data_internal,.-__svml_satan2_data_internal
> diff --git a/sysdeps/x86_64/fpu/svml_d_atan22_core.S b/sysdeps/x86_64/fpu/svml_d_atan22_core.S
> new file mode 100644
> index 0000000000..f3089e70f9
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_d_atan22_core.S
> @@ -0,0 +1,29 @@
> +/* Function atan2 vectorized with SSE2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_d_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVbN2vv_atan2)
> +WRAPPER_IMPL_SSE2_ff atan2
> +END (_ZGVbN2vv_atan2)
> +
> +#ifndef USE_MULTIARCH
> + libmvec_hidden_def (_ZGVbN2vv_atan2)
> +#endif
> diff --git a/sysdeps/x86_64/fpu/svml_d_atan24_core.S b/sysdeps/x86_64/fpu/svml_d_atan24_core.S
> new file mode 100644
> index 0000000000..8a163d12d2
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_d_atan24_core.S
> @@ -0,0 +1,29 @@
> +/* Function atan2 vectorized with AVX2, wrapper version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_d_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVdN4vv_atan2)
> +WRAPPER_IMPL_AVX_ff _ZGVbN2vv_atan2
> +END (_ZGVdN4vv_atan2)
> +
> +#ifndef USE_MULTIARCH
> + libmvec_hidden_def (_ZGVdN4vv_atan2)
> +#endif
> diff --git a/sysdeps/x86_64/fpu/svml_d_atan24_core_avx.S b/sysdeps/x86_64/fpu/svml_d_atan24_core_avx.S
> new file mode 100644
> index 0000000000..0ee5ae8faf
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_d_atan24_core_avx.S
> @@ -0,0 +1,25 @@
> +/* Function atan2 vectorized in AVX ISA as wrapper to SSE4 ISA version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_d_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVcN4vv_atan2)
> +WRAPPER_IMPL_AVX_ff _ZGVbN2vv_atan2
> +END (_ZGVcN4vv_atan2)
> diff --git a/sysdeps/x86_64/fpu/svml_d_atan28_core.S b/sysdeps/x86_64/fpu/svml_d_atan28_core.S
> new file mode 100644
> index 0000000000..b85f696686
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_d_atan28_core.S
> @@ -0,0 +1,25 @@
> +/* Function atan2 vectorized with AVX-512. Wrapper to AVX2 version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_d_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVeN8vv_atan2)
> +WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_atan2
> +END (_ZGVeN8vv_atan2)
> diff --git a/sysdeps/x86_64/fpu/svml_s_atan2f16_core.S b/sysdeps/x86_64/fpu/svml_s_atan2f16_core.S
> new file mode 100644
> index 0000000000..25acb31dfb
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_s_atan2f16_core.S
> @@ -0,0 +1,25 @@
> +/* Function atan2f vectorized with AVX-512. Wrapper to AVX2 version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_s_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVeN16vv_atan2f)
> +WRAPPER_IMPL_AVX512_ff _ZGVdN8vv_atan2f
> +END (_ZGVeN16vv_atan2f)
> diff --git a/sysdeps/x86_64/fpu/svml_s_atan2f4_core.S b/sysdeps/x86_64/fpu/svml_s_atan2f4_core.S
> new file mode 100644
> index 0000000000..bc99f0ba10
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_s_atan2f4_core.S
> @@ -0,0 +1,29 @@
> +/* Function atan2f vectorized with SSE2.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_s_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVbN4vv_atan2f)
> +WRAPPER_IMPL_SSE2_ff atan2f
> +END (_ZGVbN4vv_atan2f)
> +
> +#ifndef USE_MULTIARCH
> + libmvec_hidden_def (_ZGVbN4vv_atan2f)
> +#endif
> diff --git a/sysdeps/x86_64/fpu/svml_s_atan2f8_core.S b/sysdeps/x86_64/fpu/svml_s_atan2f8_core.S
> new file mode 100644
> index 0000000000..bfcdb3c372
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_s_atan2f8_core.S
> @@ -0,0 +1,29 @@
> +/* Function atan2f vectorized with AVX2, wrapper version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_s_wrapper_impl.h"
> +
> + .text
> +ENTRY (_ZGVdN8vv_atan2f)
> +WRAPPER_IMPL_AVX_ff _ZGVbN4vv_atan2f
> +END (_ZGVdN8vv_atan2f)
> +
> +#ifndef USE_MULTIARCH
> + libmvec_hidden_def (_ZGVdN8vv_atan2f)
> +#endif
> diff --git a/sysdeps/x86_64/fpu/svml_s_atan2f8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_atan2f8_core_avx.S
> new file mode 100644
> index 0000000000..1aa8d05822
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/svml_s_atan2f8_core_avx.S
> @@ -0,0 +1,25 @@
> +/* Function atan2f vectorized in AVX ISA as wrapper to SSE4 ISA version.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +#include "svml_s_wrapper_impl.h"
> +
> + .text
> +ENTRY(_ZGVcN8vv_atan2f)
> +WRAPPER_IMPL_AVX_ff _ZGVbN4vv_atan2f
> +END(_ZGVcN8vv_atan2f)
> diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx.c
> new file mode 100644
> index 0000000000..e423bce25b
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx.c
> @@ -0,0 +1 @@
> +#include "test-double-libmvec-atan2.c"
> diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx2.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx2.c
> new file mode 100644
> index 0000000000..e423bce25b
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx2.c
> @@ -0,0 +1 @@
> +#include "test-double-libmvec-atan2.c"
> diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx512f.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx512f.c
> new file mode 100644
> index 0000000000..e423bce25b
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan2-avx512f.c
> @@ -0,0 +1 @@
> +#include "test-double-libmvec-atan2.c"
> diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan2.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan2.c
> new file mode 100644
> index 0000000000..d0aa626d95
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan2.c
> @@ -0,0 +1,3 @@
> +#define LIBMVEC_TYPE double
> +#define LIBMVEC_FUNC atan2
> +#include "test-vector-abi-arg2.h"
> diff --git a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c
> index b1981ac7e4..37a7a1c777 100644
> --- a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVbN2v_cosh)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVbN2v_expm1)
> VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVbN2v_sinh)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrt), _ZGVbN2v_cbrt)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2), _ZGVbN2vv_atan2)
>
> #define VEC_INT_TYPE __m128i
>
> diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c
> index 47915a7e59..4313f67e06 100644
> --- a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c
> @@ -40,6 +40,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVdN4v_cosh)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVdN4v_expm1)
> VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVdN4v_sinh)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrt), _ZGVdN4v_cbrt)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2), _ZGVdN4vv_atan2)
>
> #ifndef __ILP32__
> # define VEC_INT_TYPE __m256i
> diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c
> index 5cd5049807..4b8b00f16d 100644
> --- a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVcN4v_cosh)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVcN4v_expm1)
> VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVcN4v_sinh)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrt), _ZGVcN4v_cbrt)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2), _ZGVcN4vv_atan2)
>
> #define VEC_INT_TYPE __m128i
>
> diff --git a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c
> index 83970739ab..d06522a407 100644
> --- a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVeN8v_cosh)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVeN8v_expm1)
> VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVeN8v_sinh)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrt), _ZGVeN8v_cbrt)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2), _ZGVeN8vv_atan2)
>
> #ifndef __ILP32__
> # define VEC_INT_TYPE __m512i
> diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx.c b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx.c
> new file mode 100644
> index 0000000000..5c7e2c9ad5
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx.c
> @@ -0,0 +1 @@
> +#include "test-float-libmvec-atan2f.c"
> diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx2.c b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx2.c
> new file mode 100644
> index 0000000000..5c7e2c9ad5
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx2.c
> @@ -0,0 +1 @@
> +#include "test-float-libmvec-atan2f.c"
> diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx512f.c b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx512f.c
> new file mode 100644
> index 0000000000..5c7e2c9ad5
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f-avx512f.c
> @@ -0,0 +1 @@
> +#include "test-float-libmvec-atan2f.c"
> diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atan2f.c b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f.c
> new file mode 100644
> index 0000000000..beb5c745cb
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atan2f.c
> @@ -0,0 +1,3 @@
> +#define LIBMVEC_TYPE float
> +#define LIBMVEC_FUNC atan2f
> +#include "test-vector-abi-arg2.h"
> diff --git a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c
> index 0420f11c28..0bd631bf9a 100644
> --- a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVeN16v_coshf)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVeN16v_expm1f)
> VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVeN16v_sinhf)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrtf), _ZGVeN16v_cbrtf)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2f), _ZGVeN16vv_atan2f)
>
> #define VEC_INT_TYPE __m512i
>
> diff --git a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c
> index c8f7580265..1018398bd3 100644
> --- a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVbN4v_coshf)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVbN4v_expm1f)
> VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVbN4v_sinhf)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrtf), _ZGVbN4v_cbrtf)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2f), _ZGVbN4vv_atan2f)
>
> #define VEC_INT_TYPE __m128i
>
> diff --git a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c
> index b581796b88..42ea28f30f 100644
> --- a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c
> @@ -40,6 +40,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVdN8v_coshf)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVdN8v_expm1f)
> VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVdN8v_sinhf)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrtf), _ZGVdN8v_cbrtf)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2f), _ZGVdN8vv_atan2f)
>
> /* Redefinition of wrapper to be compatible with _ZGVdN8vvv_sincosf. */
> #undef VECTOR_WRAPPER_fFF
> diff --git a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c
> index f16789e5ff..70a0216a07 100644
> --- a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c
> +++ b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c
> @@ -37,6 +37,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVcN8v_coshf)
> VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVcN8v_expm1f)
> VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVcN8v_sinhf)
> VECTOR_WRAPPER (WRAPPER_NAME (cbrtf), _ZGVcN8v_cbrtf)
> +VECTOR_WRAPPER_ff (WRAPPER_NAME (atan2f), _ZGVcN8vv_atan2f)
>
> #define VEC_INT_TYPE __m128i
>
> --
> 2.31.1
>
LGTM.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Thanks.
H.J.
next prev parent reply other threads:[~2021-12-29 21:26 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-12-29 6:39 [PATCH v5 00/18] x86-64: Add vector math functions " Sunil K Pandey
2021-12-29 6:39 ` [PATCH v5 01/18] x86-64: Add vector atan/atanf implementation " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 02/18] x86-64: Add vector asin/asinf " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 03/18] x86-64: Add vector hypot/hypotf " Sunil K Pandey
2021-12-29 21:24 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 04/18] x86-64: Add vector exp2/exp2f " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 05/18] x86-64: Add vector exp10/exp10f " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 06/18] x86-64: Add vector cosh/coshf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 07/18] x86-64: Add vector expm1/expm1f " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 08/18] x86-64: Add vector sinh/sinhf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 09/18] x86-64: Add vector cbrt/cbrtf " Sunil K Pandey
2021-12-29 21:25 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 10/18] x86-64: Add vector atan2/atan2f " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu [this message]
2021-12-29 6:39 ` [PATCH v5 11/18] x86-64: Add vector log10/log10f " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 12/18] x86-64: Add vector log2/log2f " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 13/18] x86-64: Add vector log1p/log1pf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 23:28 ` Noah Goldstein
2021-12-30 0:32 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 14/18] x86-64: Add vector atanh/atanhf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 15/18] x86-64: Add vector acosh/acoshf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 16/18] x86-64: Add vector erf/erff " Sunil K Pandey
2021-12-29 21:27 ` H.J. Lu
2021-12-29 6:39 ` [PATCH v5 17/18] x86-64: Add vector tanh/tanhf " Sunil K Pandey
2021-12-29 21:26 ` H.J. Lu
2022-01-29 1:33 ` Noah Goldstein
2021-12-29 6:40 ` [PATCH v5 18/18] x86-64: Add vector asinh/asinhf " Sunil K Pandey
2021-12-29 21:27 ` H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=YczSgo6m2tvatLYO@gmail.com \
--to=hjl.tools@gmail.com \
--cc=andrey.kolesov@intel.com \
--cc=libc-alpha@sourceware.org \
--cc=marius.cornea@intel.com \
--cc=skpgkp2@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).