From aa1809a1dfde88e5df73edba14b30e488b267343 Mon Sep 17 00:00:00 2001 From: Sunil K Pandey Date: Wed, 29 Dec 2021 09:05:18 -0800 Subject: [PATCH] x86-64: Add vector sinh/sinhf implementation to libmvec Implement vectorized sinh/sinhf containing SSE, AVX, AVX2 and AVX512 versions for libmvec as per vector ABI. It also contains accuracy and ABI tests for vector sinh/sinhf with regenerated ulps. Reviewed-by: H.J. Lu --- 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_sinh2_core-sse2.S | 20 + .../x86_64/fpu/multiarch/svml_d_sinh2_core.c | 27 + .../fpu/multiarch/svml_d_sinh2_core_sse4.S | 456 +++++++++++++++++ .../fpu/multiarch/svml_d_sinh4_core-sse.S | 20 + .../x86_64/fpu/multiarch/svml_d_sinh4_core.c | 27 + .../fpu/multiarch/svml_d_sinh4_core_avx2.S | 470 ++++++++++++++++++ .../fpu/multiarch/svml_d_sinh8_core-avx2.S | 20 + .../x86_64/fpu/multiarch/svml_d_sinh8_core.c | 27 + .../fpu/multiarch/svml_d_sinh8_core_avx512.S | 461 +++++++++++++++++ .../fpu/multiarch/svml_s_sinhf16_core-avx2.S | 20 + .../fpu/multiarch/svml_s_sinhf16_core.c | 28 ++ .../multiarch/svml_s_sinhf16_core_avx512.S | 318 ++++++++++++ .../fpu/multiarch/svml_s_sinhf4_core-sse2.S | 20 + .../x86_64/fpu/multiarch/svml_s_sinhf4_core.c | 28 ++ .../fpu/multiarch/svml_s_sinhf4_core_sse4.S | 308 ++++++++++++ .../fpu/multiarch/svml_s_sinhf8_core-sse.S | 20 + .../x86_64/fpu/multiarch/svml_s_sinhf8_core.c | 28 ++ .../fpu/multiarch/svml_s_sinhf8_core_avx2.S | 309 ++++++++++++ sysdeps/x86_64/fpu/svml_d_sinh2_core.S | 29 ++ sysdeps/x86_64/fpu/svml_d_sinh4_core.S | 29 ++ sysdeps/x86_64/fpu/svml_d_sinh4_core_avx.S | 25 + sysdeps/x86_64/fpu/svml_d_sinh8_core.S | 25 + sysdeps/x86_64/fpu/svml_s_sinhf16_core.S | 25 + sysdeps/x86_64/fpu/svml_s_sinhf4_core.S | 29 ++ sysdeps/x86_64/fpu/svml_s_sinhf8_core.S | 29 ++ sysdeps/x86_64/fpu/svml_s_sinhf8_core_avx.S | 25 + .../x86_64/fpu/test-double-libmvec-sinh-avx.c | 1 + .../fpu/test-double-libmvec-sinh-avx2.c | 1 + .../fpu/test-double-libmvec-sinh-avx512f.c | 1 + sysdeps/x86_64/fpu/test-double-libmvec-sinh.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 + .../x86_64/fpu/test-float-libmvec-sinhf-avx.c | 1 + .../fpu/test-float-libmvec-sinhf-avx2.c | 1 + .../fpu/test-float-libmvec-sinhf-avx512f.c | 1 + sysdeps/x86_64/fpu/test-float-libmvec-sinhf.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, 2894 insertions(+), 1 deletion(-) create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core-sse2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core_sse4.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core-sse.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core_avx2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core-avx2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core_avx512.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core-avx2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core_avx512.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core-sse2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core_sse4.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core-sse.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core.c create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core_avx2.S create mode 100644 sysdeps/x86_64/fpu/svml_d_sinh2_core.S create mode 100644 sysdeps/x86_64/fpu/svml_d_sinh4_core.S create mode 100644 sysdeps/x86_64/fpu/svml_d_sinh4_core_avx.S create mode 100644 sysdeps/x86_64/fpu/svml_d_sinh8_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_sinhf16_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_sinhf4_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_sinhf8_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_sinhf8_core_avx.S create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx.c create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx2.c create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx512f.c create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-sinh.c create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx.c create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx2.c create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx512f.c create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-sinhf.c diff --git a/bits/libm-simd-decl-stubs.h b/bits/libm-simd-decl-stubs.h index 28dc4a82c5..6347320521 100644 --- a/bits/libm-simd-decl-stubs.h +++ b/bits/libm-simd-decl-stubs.h @@ -186,4 +186,15 @@ #define __DECL_SIMD_expm1f32x #define __DECL_SIMD_expm1f64x #define __DECL_SIMD_expm1f128x + +#define __DECL_SIMD_sinh +#define __DECL_SIMD_sinhf +#define __DECL_SIMD_sinhl +#define __DECL_SIMD_sinhf16 +#define __DECL_SIMD_sinhf32 +#define __DECL_SIMD_sinhf64 +#define __DECL_SIMD_sinhf128 +#define __DECL_SIMD_sinhf32x +#define __DECL_SIMD_sinhf64x +#define __DECL_SIMD_sinhf128x #endif diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h index c57adc8ace..673b3a93ba 100644 --- a/math/bits/mathcalls.h +++ b/math/bits/mathcalls.h @@ -70,7 +70,7 @@ __MATHCALL (tan,, (_Mdouble_ __x)); /* Hyperbolic cosine of X. */ __MATHCALL_VEC (cosh,, (_Mdouble_ __x)); /* Hyperbolic sine of X. */ -__MATHCALL (sinh,, (_Mdouble_ __x)); +__MATHCALL_VEC (sinh,, (_Mdouble_ __x)); /* Hyperbolic tangent of X. */ __MATHCALL (tanh,, (_Mdouble_ __x)); diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist index c9d3213bd3..f9d7b085ab 100644 --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist @@ -53,6 +53,7 @@ GLIBC_2.35 _ZGVbN2v_cosh F 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_hypot F GLIBC_2.35 _ZGVbN4v_acosf F GLIBC_2.35 _ZGVbN4v_asinf F @@ -61,6 +62,7 @@ GLIBC_2.35 _ZGVbN4v_coshf F 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_hypotf F GLIBC_2.35 _ZGVcN4v_acos F GLIBC_2.35 _ZGVcN4v_asin F @@ -69,6 +71,7 @@ GLIBC_2.35 _ZGVcN4v_cosh F 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_hypot F GLIBC_2.35 _ZGVcN8v_acosf F GLIBC_2.35 _ZGVcN8v_asinf F @@ -77,6 +80,7 @@ GLIBC_2.35 _ZGVcN8v_coshf F 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_hypotf F GLIBC_2.35 _ZGVdN4v_acos F GLIBC_2.35 _ZGVdN4v_asin F @@ -85,6 +89,7 @@ GLIBC_2.35 _ZGVdN4v_cosh F 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_hypot F GLIBC_2.35 _ZGVdN8v_acosf F GLIBC_2.35 _ZGVdN8v_asinf F @@ -93,6 +98,7 @@ GLIBC_2.35 _ZGVdN8v_coshf F 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_hypotf F GLIBC_2.35 _ZGVeN16v_acosf F GLIBC_2.35 _ZGVeN16v_asinf F @@ -101,6 +107,7 @@ GLIBC_2.35 _ZGVeN16v_coshf F 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_hypotf F GLIBC_2.35 _ZGVeN8v_acos F GLIBC_2.35 _ZGVeN8v_asin F @@ -109,4 +116,5 @@ GLIBC_2.35 _ZGVeN8v_cosh F 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_hypot F diff --git a/sysdeps/x86/fpu/bits/math-vector.h b/sysdeps/x86/fpu/bits/math-vector.h index e2f98e176f..51a41cfebc 100644 --- a/sysdeps/x86/fpu/bits/math-vector.h +++ b/sysdeps/x86/fpu/bits/math-vector.h @@ -90,6 +90,10 @@ # define __DECL_SIMD_expm1 __DECL_SIMD_x86_64 # undef __DECL_SIMD_expm1f # define __DECL_SIMD_expm1f __DECL_SIMD_x86_64 +# undef __DECL_SIMD_sinh +# define __DECL_SIMD_sinh __DECL_SIMD_x86_64 +# undef __DECL_SIMD_sinhf +# define __DECL_SIMD_sinhf __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 43233059f6..91e9b4fc83 100644 --- a/sysdeps/x86/fpu/finclude/math-vector-fortran.h +++ b/sysdeps/x86/fpu/finclude/math-vector-fortran.h @@ -44,6 +44,8 @@ !GCC$ builtin (coshf) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (expm1) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (expm1f) attributes simd (notinbranch) if('x86_64') +!GCC$ builtin (sinh) attributes simd (notinbranch) if('x86_64') +!GCC$ builtin (sinhf) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (cos) attributes simd (notinbranch) if('x32') !GCC$ builtin (cosf) attributes simd (notinbranch) if('x32') @@ -73,3 +75,5 @@ !GCC$ builtin (coshf) attributes simd (notinbranch) if('x32') !GCC$ builtin (expm1) attributes simd (notinbranch) if('x32') !GCC$ builtin (expm1f) attributes simd (notinbranch) if('x32') +!GCC$ builtin (sinh) attributes simd (notinbranch) if('x32') +!GCC$ builtin (sinhf) attributes simd (notinbranch) if('x32') diff --git a/sysdeps/x86_64/fpu/Makeconfig b/sysdeps/x86_64/fpu/Makeconfig index 8de8214971..81e9fc95b2 100644 --- a/sysdeps/x86_64/fpu/Makeconfig +++ b/sysdeps/x86_64/fpu/Makeconfig @@ -36,6 +36,7 @@ libmvec-funcs = \ pow \ sin \ sincos \ + sinh \ # Define libmvec function for benchtests directory. libmvec-bench-funcs = \ diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions index 58debb2dbe..2710446d12 100644 --- a/sysdeps/x86_64/fpu/Versions +++ b/sysdeps/x86_64/fpu/Versions @@ -21,6 +21,7 @@ libmvec { _ZGVbN2v_exp10; _ZGVcN4v_exp10; _ZGVdN4v_exp10; _ZGVeN8v_exp10; _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_hypot; _ZGVcN4vv_hypot; _ZGVdN4vv_hypot; _ZGVeN8vv_hypot; _ZGVbN4v_acosf; _ZGVcN8v_acosf; _ZGVdN8v_acosf; _ZGVeN16v_acosf; _ZGVbN4v_asinf; _ZGVcN8v_asinf; _ZGVdN8v_asinf; _ZGVeN16v_asinf; @@ -29,6 +30,7 @@ libmvec { _ZGVbN4v_exp10f; _ZGVcN8v_exp10f; _ZGVdN8v_exp10f; _ZGVeN16v_exp10f; _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_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 f05ece8c8a..f4b313119d 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -1840,6 +1840,26 @@ float: 3 float128: 4 ldouble: 5 +Function: "sinh_vlen16": +float: 1 + +Function: "sinh_vlen2": +double: 2 + +Function: "sinh_vlen4": +double: 2 +float: 1 + +Function: "sinh_vlen4_avx2": +double: 2 + +Function: "sinh_vlen8": +double: 2 +float: 1 + +Function: "sinh_vlen8_avx2": +float: 1 + Function: "tan": float: 1 float128: 1 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core-sse2.S new file mode 100644 index 0000000000..ca12ad6678 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core-sse2.S @@ -0,0 +1,20 @@ +/* SSE2 version of vectorized sinh, 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 + . */ + +#define _ZGVbN2v_sinh _ZGVbN2v_sinh_sse2 +#include "../svml_d_sinh2_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core.c new file mode 100644 index 0000000000..c0344b2902 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized sinh, 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 + . */ + +#define SYMBOL_NAME _ZGVbN2v_sinh +#include "ifunc-mathvec-sse4_1.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVbN2v_sinh, __GI__ZGVbN2v_sinh, __redirect__ZGVbN2v_sinh) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core_sse4.S new file mode 100644 index 0000000000..80d19e9dba --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh2_core_sse4.S @@ -0,0 +1,456 @@ +/* Function sinh 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_dsinh_data_internal + */ +#define _dbInvLn2 0 +#define _dbLn2hi 16 +#define _dbLn2lo 32 +#define _dSign 48 +#define _dbT 64 +#define _dbShifter 2112 +#define _iDomainRange 2128 +#define _dPC2 2144 +#define _dPC3 2160 +#define _dPC4 2176 +#define _dPC5 2192 +#define _lIndexMask 2208 + +#include + + .text + .section .text.sse4,"ax",@progbits +ENTRY(_ZGVbN2v_sinh_sse4) + subq $72, %rsp + cfi_def_cfa_offset(80) + movaps %xmm0, %xmm2 + +/* Abs argument */ + movups _dSign+__svml_dsinh_data_internal(%rip), %xmm0 + lea _dbT+8+__svml_dsinh_data_internal(%rip), %rsi + andps %xmm2, %xmm0 + movaps %xmm0, %xmm1 + +/* + * Load argument + * dM = x*2^K/log(2) + RShifter + */ + movups _dbInvLn2+__svml_dsinh_data_internal(%rip), %xmm10 + pxor %xmm2, %xmm1 + mulpd %xmm1, %xmm10 + movups _dbShifter+__svml_dsinh_data_internal(%rip), %xmm5 + addpd %xmm5, %xmm10 + +/* + * R + * dN = dM - RShifter + */ + movaps %xmm10, %xmm7 + subpd %xmm5, %xmm7 + +/* dR = dX - dN*Log2_hi/2^K */ + movups _dbLn2hi+__svml_dsinh_data_internal(%rip), %xmm6 + mulpd %xmm7, %xmm6 + +/* dR = (dX - dN*Log2_hi/2^K) - dN*Log2_lo/2^K */ + movups _dbLn2lo+__svml_dsinh_data_internal(%rip), %xmm8 + mulpd %xmm7, %xmm8 + +/* + * Check for overflow\underflow + * + */ + pshufd $221, %xmm1, %xmm4 + subpd %xmm6, %xmm1 + subpd %xmm8, %xmm1 + +/* VLOAD_CONST( D, dPC[0], TAB._dPC1 ); */ + movq _iDomainRange+__svml_dsinh_data_internal(%rip), %xmm3 + pcmpgtd %xmm3, %xmm4 + +/* dR2 = dR^2 */ + movaps %xmm1, %xmm3 + mulpd %xmm1, %xmm3 + movmskps %xmm4, %edx + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*a5)) = r + r*(r^2*(a3+r^2*a5)) .... + * dSinh_r = (a3+r^2*a5) + */ + movups _dPC5+__svml_dsinh_data_internal(%rip), %xmm12 + +/* + * poly(r) = (dG2+dG1)+dG3*sinh(dR)+dG1*sinh(dR)+(dG1+dG2)*dR2*(a2 +a4*dR2) + * dOut = (a2 +a4*dR2) + */ + movups _dPC4+__svml_dsinh_data_internal(%rip), %xmm13 + mulpd %xmm3, %xmm12 + mulpd %xmm3, %xmm13 + addpd _dPC3+__svml_dsinh_data_internal(%rip), %xmm12 + addpd _dPC2+__svml_dsinh_data_internal(%rip), %xmm13 + +/* dSinh_r = r^2*(a3+r^2*a5) */ + mulpd %xmm3, %xmm12 + +/* dOut = dR2*(a2 +a4*dR2) */ + mulpd %xmm13, %xmm3 + +/* dSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + mulpd %xmm1, %xmm12 + +/* + * Index and lookup + * j + */ + movups _lIndexMask+__svml_dsinh_data_internal(%rip), %xmm9 + andps %xmm10, %xmm9 + movd %xmm9, %eax + +/* split j and N */ + pxor %xmm9, %xmm10 + +/* + * G1,G2,G3: dTdif,dTn * 2^N,2^(-N) + * lM now is an EXP(2^N) + */ + psllq $45, %xmm10 + +/* */ + movaps %xmm10, %xmm4 + pextrw $4, %xmm9, %ecx + addpd %xmm12, %xmm1 + shll $4, %eax + shll $4, %ecx + movq (%rax,%rsi), %xmm11 + movhpd (%rcx,%rsi), %xmm11 + paddq %xmm11, %xmm4 + +/* */ + psubq %xmm10, %xmm11 + +/* dG3 = dTn*2^N + dTn*2^-N */ + movdqa %xmm4, %xmm14 + addpd %xmm11, %xmm14 + +/* dG2 = dTn*2^N - dTn*2^-N */ + subpd %xmm11, %xmm4 + movq -8(%rax,%rsi), %xmm15 + movhpd -8(%rcx,%rsi), %xmm15 + paddq %xmm10, %xmm15 + +/* dG2 += dG1 */ + addpd %xmm15, %xmm4 + +/* dG1 += dG3 */ + addpd %xmm14, %xmm15 + +/* dOut = dG2*dR2*(a2 +a4*dR2) */ + mulpd %xmm4, %xmm3 + +/* dOut = dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + mulpd %xmm15, %xmm1 + addpd %xmm1, %xmm3 + +/* dOut = dG2 + dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + addpd %xmm3, %xmm4 + +/* Ret H */ + orps %xmm4, %xmm0 + andl $3, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm2 + +/* Restore registers + * and exit the function + */ + +L(EXIT): + addq $72, %rsp + cfi_def_cfa_offset(8) + ret + cfi_def_cfa_offset(80) + +/* Branch to process + * special inputs + */ + +L(SPECIAL_VALUES_BRANCH): + movups %xmm2, 32(%rsp) + movups %xmm0, 48(%rsp) + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 + + xorl %eax, %eax + movq %r12, 16(%rsp) + cfi_offset(12, -64) + movl %eax, %r12d + movq %r13, 8(%rsp) + cfi_offset(13, -72) + movl %edx, %r13d + movq %r14, (%rsp) + cfi_offset(14, -80) + # 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 48(%rsp), %xmm0 + +/* Go to exit */ + jmp L(EXIT) + cfi_offset(12, -64) + cfi_offset(13, -72) + cfi_offset(14, -80) + # 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 + call sinh@PLT + # LOE rbx rbp r14 r15 r12d r13d xmm0 + + movsd %xmm0, 48(%rsp,%r14,8) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx rbp r15 r12d r13d +END(_ZGVbN2v_sinh_sse4) + + .section .rodata, "a" + .align 16 + +#ifdef __svml_dsinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(16)) VUINT32 _dbInvLn2[2][2]; + __declspec(align(16)) VUINT32 _dbLn2hi[2][2]; + __declspec(align(16)) VUINT32 _dbLn2lo[2][2]; + __declspec(align(16)) VUINT32 _dSign[2][2]; //0x8000000000000000 + __declspec(align(16)) VUINT32 _dbT[(1<<7)][2][2]; //precalc poly coeff + __declspec(align(16)) VUINT32 _dbShifter[2][2]; + __declspec(align(16)) VUINT32 _iDomainRange[4][1]; + __declspec(align(16)) VUINT32 _dPC2[2][2]; + __declspec(align(16)) VUINT32 _dPC3[2][2]; + __declspec(align(16)) VUINT32 _dPC4[2][2]; + __declspec(align(16)) VUINT32 _dPC5[2][2]; + __declspec(align(16)) VUINT32 _lIndexMask[2][2]; +} __svml_dsinh_data_internal; +#endif +__svml_dsinh_data_internal: + .quad 0x3FF71547652B82FE, 0x3FF71547652B82FE /* _dbInvLn2 = 1/log(2) */ + .align 16 + .quad 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000 /* _dbLn2hi = log(2) hi*/ + .align 16 + .quad 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A /* _dbLn2lo = log(2) lo*/ + .align 16 + .quad 0x8000000000000000, 0x8000000000000000 /* _dSign */ + //_dbT + .align 16 + .quad 0x0000000000000000, 0x3FE0000000000000 //2^( 0 /128-1) - 2^(- 0 /128-1), 2^(- 0 /128-1) + .quad 0x3F762E4A19BD1E74, 0x3FDFD3C22B8F71F1 //2^( 1 /128-1) - 2^(- 1 /128-1), 2^(- 1 /128-1) + .quad 0x3F862E5F6A0DFD36, 0x3FDFA7C1819E90D8 //2^( 2 /128-1) - 2^(- 2 /128-1), 2^(- 2 /128-1) + .quad 0x3F90A2E234040F5F, 0x3FDF7BFDAD9CBE14 //2^( 3 /128-1) - 2^(- 3 /128-1), 2^(- 3 /128-1) + .quad 0x3F962EB4ABCC5A81, 0x3FDF50765B6E4540 //2^( 4 /128-1) - 2^(- 4 /128-1), 2^(- 4 /128-1) + .quad 0x3F9BBAB1C5033244, 0x3FDF252B376BBA97 //2^( 5 /128-1) - 2^(- 5 /128-1), 2^(- 5 /128-1) + .quad 0x3FA0A372144EEB45, 0x3FDEFA1BEE615A27 //2^( 6 /128-1) - 2^(- 6 /128-1), 2^(- 6 /128-1) + .quad 0x3FA369AB3FFBF8B0, 0x3FDECF482D8E67F1 //2^( 7 /128-1) - 2^(- 7 /128-1), 2^(- 7 /128-1) + .quad 0x3FA63009BA740A2A, 0x3FDEA4AFA2A490DA //2^( 8 /128-1) - 2^(- 8 /128-1), 2^(- 8 /128-1) + .quad 0x3FA8F692D8EA1B5A, 0x3FDE7A51FBC74C83 //2^( 9 /128-1) - 2^(- 9 /128-1), 2^(- 9 /128-1) + .quad 0x3FABBD4BF0E31A6F, 0x3FDE502EE78B3FF6 //2^( 10 /128-1) - 2^(- 10 /128-1), 2^(- 10 /128-1) + .quad 0x3FAE843A5840286A, 0x3FDE264614F5A129 //2^( 11 /128-1) - 2^(- 11 /128-1), 2^(- 11 /128-1) + .quad 0x3FB0A5B1B2A46D0A, 0x3FDDFC97337B9B5F //2^( 12 /128-1) - 2^(- 12 /128-1), 2^(- 12 /128-1) + .quad 0x3FB20966375ABCDF, 0x3FDDD321F301B460 //2^( 13 /128-1) - 2^(- 13 /128-1), 2^(- 13 /128-1) + .quad 0x3FB36D3D65DCA4E8, 0x3FDDA9E603DB3285 //2^( 14 /128-1) - 2^(- 14 /128-1), 2^(- 14 /128-1) + .quad 0x3FB4D139EA06642A, 0x3FDD80E316C98398 //2^( 15 /128-1) - 2^(- 15 /128-1), 2^(- 15 /128-1) + .quad 0x3FB6355E6FFBF9BA, 0x3FDD5818DCFBA487 //2^( 16 /128-1) - 2^(- 16 /128-1), 2^(- 16 /128-1) + .quad 0x3FB799ADA42E4788, 0x3FDD2F87080D89F2 //2^( 17 /128-1) - 2^(- 17 /128-1), 2^(- 17 /128-1) + .quad 0x3FB8FE2A336035BC, 0x3FDD072D4A07897C //2^( 18 /128-1) - 2^(- 18 /128-1), 2^(- 18 /128-1) + .quad 0x3FBA62D6CAABD6B6, 0x3FDCDF0B555DC3FA //2^( 19 /128-1) - 2^(- 19 /128-1), 2^(- 19 /128-1) + .quad 0x3FBBC7B617878BAF, 0x3FDCB720DCEF9069 //2^( 20 /128-1) - 2^(- 20 /128-1), 2^(- 20 /128-1) + .quad 0x3FBD2CCAC7CB2A11, 0x3FDC8F6D9406E7B5 //2^( 21 /128-1) - 2^(- 21 /128-1), 2^(- 21 /128-1) + .quad 0x3FBE921789B52185, 0x3FDC67F12E57D14B //2^( 22 /128-1) - 2^(- 22 /128-1), 2^(- 22 /128-1) + .quad 0x3FBFF79F0BEFA2C7, 0x3FDC40AB5FFFD07A //2^( 23 /128-1) - 2^(- 23 /128-1), 2^(- 23 /128-1) + .quad 0x3FC0AEB1FECAE3A9, 0x3FDC199BDD85529C //2^( 24 /128-1) - 2^(- 24 /128-1), 2^(- 24 /128-1) + .quad 0x3FC161B4871C5CEC, 0x3FDBF2C25BD71E09 //2^( 25 /128-1) - 2^(- 25 /128-1), 2^(- 25 /128-1) + .quad 0x3FC214D876F26FD0, 0x3FDBCC1E904BC1D2 //2^( 26 /128-1) - 2^(- 26 /128-1), 2^(- 26 /128-1) + .quad 0x3FC2C81F2693816F, 0x3FDBA5B030A1064A //2^( 27 /128-1) - 2^(- 27 /128-1), 2^(- 27 /128-1) + .quad 0x3FC37B89EE88BEF7, 0x3FDB7F76F2FB5E47 //2^( 28 /128-1) - 2^(- 28 /128-1), 2^(- 28 /128-1) + .quad 0x3FC42F1A27A0B3CD, 0x3FDB59728DE5593A //2^( 29 /128-1) - 2^(- 29 /128-1), 2^(- 29 /128-1) + .quad 0x3FC4E2D12AF1E037, 0x3FDB33A2B84F15FB //2^( 30 /128-1) - 2^(- 30 /128-1), 2^(- 30 /128-1) + .quad 0x3FC596B051DD508D, 0x3FDB0E07298DB666 //2^( 31 /128-1) - 2^(- 31 /128-1), 2^(- 31 /128-1) + .quad 0x3FC64AB8F61134FA, 0x3FDAE89F995AD3AD //2^( 32 /128-1) - 2^(- 32 /128-1), 2^(- 32 /128-1) + .quad 0x3FC6FEEC718B79D1, 0x3FDAC36BBFD3F37A //2^( 33 /128-1) - 2^(- 33 /128-1), 2^(- 33 /128-1) + .quad 0x3FC7B34C1E9C607F, 0x3FDA9E6B5579FDBF //2^( 34 /128-1) - 2^(- 34 /128-1), 2^(- 34 /128-1) + .quad 0x3FC867D957E91912, 0x3FDA799E1330B358 //2^( 35 /128-1) - 2^(- 35 /128-1), 2^(- 35 /128-1) + .quad 0x3FC91C95786E5C72, 0x3FDA5503B23E255D //2^( 36 /128-1) - 2^(- 36 /128-1), 2^(- 36 /128-1) + .quad 0x3FC9D181DB83072F, 0x3FDA309BEC4A2D33 //2^( 37 /128-1) - 2^(- 37 /128-1), 2^(- 37 /128-1) + .quad 0x3FCA869FDCDAB512, 0x3FDA0C667B5DE565 //2^( 38 /128-1) - 2^(- 38 /128-1), 2^(- 38 /128-1) + .quad 0x3FCB3BF0D8885D4C, 0x3FD9E86319E32323 //2^( 39 /128-1) - 2^(- 39 /128-1), 2^(- 39 /128-1) + .quad 0x3FCBF1762B00EF69, 0x3FD9C49182A3F090 //2^( 40 /128-1) - 2^(- 40 /128-1), 2^(- 40 /128-1) + .quad 0x3FCCA731311DF0FB, 0x3FD9A0F170CA07BA //2^( 41 /128-1) - 2^(- 41 /128-1), 2^(- 41 /128-1) + .quad 0x3FCD5D2348201C09, 0x3FD97D829FDE4E50 //2^( 42 /128-1) - 2^(- 42 /128-1), 2^(- 42 /128-1) + .quad 0x3FCE134DCDB1FE3E, 0x3FD95A44CBC8520F //2^( 43 /128-1) - 2^(- 43 /128-1), 2^(- 43 /128-1) + .quad 0x3FCEC9B21FEA98EA, 0x3FD93737B0CDC5E5 //2^( 44 /128-1) - 2^(- 44 /128-1), 2^(- 44 /128-1) + .quad 0x3FCF80519D5001D3, 0x3FD9145B0B91FFC6 //2^( 45 /128-1) - 2^(- 45 /128-1), 2^(- 45 /128-1) + .quad 0x3FD01B96D26D026A, 0x3FD8F1AE99157736 //2^( 46 /128-1) - 2^(- 46 /128-1), 2^(- 46 /128-1) + .quad 0x3FD07723CAFA6331, 0x3FD8CF3216B5448C //2^( 47 /128-1) - 2^(- 47 /128-1), 2^(- 47 /128-1) + .quad 0x3FD0D2D06841B373, 0x3FD8ACE5422AA0DB //2^( 48 /128-1) - 2^(- 48 /128-1), 2^(- 48 /128-1) + .quad 0x3FD12E9D5A715381, 0x3FD88AC7D98A6699 //2^( 49 /128-1) - 2^(- 49 /128-1), 2^(- 49 /128-1) + .quad 0x3FD18A8B51F5C661, 0x3FD868D99B4492ED //2^( 50 /128-1) - 2^(- 50 /128-1), 2^(- 50 /128-1) + .quad 0x3FD1E69AFF7B04D7, 0x3FD8471A4623C7AD //2^( 51 /128-1) - 2^(- 51 /128-1), 2^(- 51 /128-1) + .quad 0x3FD242CD13EDD0F1, 0x3FD82589994CCE13 //2^( 52 /128-1) - 2^(- 52 /128-1), 2^(- 52 /128-1) + .quad 0x3FD29F22407D0A0C, 0x3FD80427543E1A12 //2^( 53 /128-1) - 2^(- 53 /128-1), 2^(- 53 /128-1) + .quad 0x3FD2FB9B369B0153, 0x3FD7E2F336CF4E62 //2^( 54 /128-1) - 2^(- 54 /128-1), 2^(- 54 /128-1) + .quad 0x3FD35838A7FECEC8, 0x3FD7C1ED0130C132 //2^( 55 /128-1) - 2^(- 55 /128-1), 2^(- 55 /128-1) + .quad 0x3FD3B4FB46A5A6CC, 0x3FD7A11473EB0187 //2^( 56 /128-1) - 2^(- 56 /128-1), 2^(- 56 /128-1) + .quad 0x3FD411E3C4D4302F, 0x3FD780694FDE5D3F //2^( 57 /128-1) - 2^(- 57 /128-1), 2^(- 57 /128-1) + .quad 0x3FD46EF2D517DAC8, 0x3FD75FEB564267C9 //2^( 58 /128-1) - 2^(- 58 /128-1), 2^(- 58 /128-1) + .quad 0x3FD4CC292A48369E, 0x3FD73F9A48A58174 //2^( 59 /128-1) - 2^(- 59 /128-1), 2^(- 59 /128-1) + .quad 0x3FD5298777884B96, 0x3FD71F75E8EC5F74 //2^( 60 /128-1) - 2^(- 60 /128-1), 2^(- 60 /128-1) + .quad 0x3FD5870E7047F1BC, 0x3FD6FF7DF9519484 //2^( 61 /128-1) - 2^(- 61 /128-1), 2^(- 61 /128-1) + .quad 0x3FD5E4BEC8452A1A, 0x3FD6DFB23C651A2F //2^( 62 /128-1) - 2^(- 62 /128-1), 2^(- 62 /128-1) + .quad 0x3FD64299338D7827, 0x3FD6C012750BDABF //2^( 63 /128-1) - 2^(- 63 /128-1), 2^(- 63 /128-1) + .quad 0x3FD6A09E667F3BCD, 0x3FD6A09E667F3BCD //2^( 64 /128-1) - 2^(- 64 /128-1), 2^(- 64 /128-1) + .quad 0x3FD6FECF15CB0C0B, 0x3FD68155D44CA973 //2^( 65 /128-1) - 2^(- 65 /128-1), 2^(- 65 /128-1) + .quad 0x3FD75D2BF6751239, 0x3FD6623882552225 //2^( 66 /128-1) - 2^(- 66 /128-1), 2^(- 66 /128-1) + .quad 0x3FD7BBB5BDD665E8, 0x3FD6434634CCC320 //2^( 67 /128-1) - 2^(- 67 /128-1), 2^(- 67 /128-1) + .quad 0x3FD81A6D219E6963, 0x3FD6247EB03A5585 //2^( 68 /128-1) - 2^(- 68 /128-1), 2^(- 68 /128-1) + .quad 0x3FD87952D7D426DF, 0x3FD605E1B976DC09 //2^( 69 /128-1) - 2^(- 69 /128-1), 2^(- 69 /128-1) + .quad 0x3FD8D86796D7AE49, 0x3FD5E76F15AD2148 //2^( 70 /128-1) - 2^(- 70 /128-1), 2^(- 70 /128-1) + .quad 0x3FD937AC156373C8, 0x3FD5C9268A5946B7 //2^( 71 /128-1) - 2^(- 71 /128-1), 2^(- 71 /128-1) + .quad 0x3FD997210A8DAEE4, 0x3FD5AB07DD485429 //2^( 72 /128-1) - 2^(- 72 /128-1), 2^(- 72 /128-1) + .quad 0x3FD9F6C72DC9BA68, 0x3FD58D12D497C7FD //2^( 73 /128-1) - 2^(- 73 /128-1), 2^(- 73 /128-1) + .quad 0x3FDA569F36E974EA, 0x3FD56F4736B527DA //2^( 74 /128-1) - 2^(- 74 /128-1), 2^(- 74 /128-1) + .quad 0x3FDAB6A9DE1EA215, 0x3FD551A4CA5D920F //2^( 75 /128-1) - 2^(- 75 /128-1), 2^(- 75 /128-1) + .quad 0x3FDB16E7DBFC4CA3, 0x3FD5342B569D4F82 //2^( 76 /128-1) - 2^(- 76 /128-1), 2^(- 76 /128-1) + .quad 0x3FDB7759E9782918, 0x3FD516DAA2CF6642 //2^( 77 /128-1) - 2^(- 77 /128-1), 2^(- 77 /128-1) + .quad 0x3FDBD800BFEBF932, 0x3FD4F9B2769D2CA7 //2^( 78 /128-1) - 2^(- 78 /128-1), 2^(- 78 /128-1) + .quad 0x3FDC38DD1916F025, 0x3FD4DCB299FDDD0D //2^( 79 /128-1) - 2^(- 79 /128-1), 2^(- 79 /128-1) + .quad 0x3FDC99EFAF1F1790, 0x3FD4BFDAD5362A27 //2^( 80 /128-1) - 2^(- 80 /128-1), 2^(- 80 /128-1) + .quad 0x3FDCFB393C92B539, 0x3FD4A32AF0D7D3DE //2^( 81 /128-1) - 2^(- 81 /128-1), 2^(- 81 /128-1) + .quad 0x3FDD5CBA7C69B19C, 0x3FD486A2B5C13CD0 //2^( 82 /128-1) - 2^(- 82 /128-1), 2^(- 82 /128-1) + .quad 0x3FDDBE742A06FF34, 0x3FD46A41ED1D0057 //2^( 83 /128-1) - 2^(- 83 /128-1), 2^(- 83 /128-1) + .quad 0x3FDE2067013A029D, 0x3FD44E086061892D //2^( 84 /128-1) - 2^(- 84 /128-1), 2^(- 84 /128-1) + .quad 0x3FDE8293BE3FFB87, 0x3FD431F5D950A897 //2^( 85 /128-1) - 2^(- 85 /128-1), 2^(- 85 /128-1) + .quad 0x3FDEE4FB1DC56E75, 0x3FD4160A21F72E2A //2^( 86 /128-1) - 2^(- 86 /128-1), 2^(- 86 /128-1) + .quad 0x3FDF479DDCE78F58, 0x3FD3FA4504AC801C //2^( 87 /128-1) - 2^(- 87 /128-1), 2^(- 87 /128-1) + .quad 0x3FDFAA7CB935ACFE, 0x3FD3DEA64C123422 //2^( 88 /128-1) - 2^(- 88 /128-1), 2^(- 88 /128-1) + .quad 0x3FE006CC38594EB1, 0x3FD3C32DC313A8E5 //2^( 89 /128-1) - 2^(- 89 /128-1), 2^(- 89 /128-1) + .quad 0x3FE03878E0EB1569, 0x3FD3A7DB34E59FF7 //2^( 90 /128-1) - 2^(- 90 /128-1), 2^(- 90 /128-1) + .quad 0x3FE06A44B5C74101, 0x3FD38CAE6D05D866 //2^( 91 /128-1) - 2^(- 91 /128-1), 2^(- 91 /128-1) + .quad 0x3FE09C3016A0D077, 0x3FD371A7373AA9CB //2^( 92 /128-1) - 2^(- 92 /128-1), 2^(- 92 /128-1) + .quad 0x3FE0CE3B63676360, 0x3FD356C55F929FF1 //2^( 93 /128-1) - 2^(- 93 /128-1), 2^(- 93 /128-1) + .quad 0x3FE10066FC47F240, 0x3FD33C08B26416FF //2^( 94 /128-1) - 2^(- 94 /128-1), 2^(- 94 /128-1) + .quad 0x3FE132B341AD8761, 0x3FD32170FC4CD831 //2^( 95 /128-1) - 2^(- 95 /128-1), 2^(- 95 /128-1) + .quad 0x3FE165209441F823, 0x3FD306FE0A31B715 //2^( 96 /128-1) - 2^(- 96 /128-1), 2^(- 96 /128-1) + .quad 0x3FE197AF54EE9EBB, 0x3FD2ECAFA93E2F56 //2^( 97 /128-1) - 2^(- 97 /128-1), 2^(- 97 /128-1) + .quad 0x3FE1CA5FE4DD1475, 0x3FD2D285A6E4030B //2^( 98 /128-1) - 2^(- 98 /128-1), 2^(- 98 /128-1) + .quad 0x3FE1FD32A577EC72, 0x3FD2B87FD0DAD990 //2^( 99 /128-1) - 2^(- 99 /128-1), 2^(- 99 /128-1) + .quad 0x3FE23027F86B6ED6, 0x3FD29E9DF51FDEE1 //2^( 100 /128-1) - 2^(- 100 /128-1), 2^(- 100 /128-1) + .quad 0x3FE263403FA65489, 0x3FD284DFE1F56381 //2^( 101 /128-1) - 2^(- 101 /128-1), 2^(- 101 /128-1) + .quad 0x3FE2967BDD5A8364, 0x3FD26B4565E27CDD //2^( 102 /128-1) - 2^(- 102 /128-1), 2^(- 102 /128-1) + .quad 0x3FE2C9DB33FDCAE9, 0x3FD251CE4FB2A63F //2^( 103 /128-1) - 2^(- 103 /128-1), 2^(- 103 /128-1) + .quad 0x3FE2FD5EA64AA180, 0x3FD2387A6E756238 //2^( 104 /128-1) - 2^(- 104 /128-1), 2^(- 104 /128-1) + .quad 0x3FE331069740E22F, 0x3FD21F49917DDC96 //2^( 105 /128-1) - 2^(- 105 /128-1), 2^(- 105 /128-1) + .quad 0x3FE364D36A268AE0, 0x3FD2063B88628CD6 //2^( 106 /128-1) - 2^(- 106 /128-1), 2^(- 106 /128-1) + .quad 0x3FE398C582887B27, 0x3FD1ED5022FCD91D //2^( 107 /128-1) - 2^(- 107 /128-1), 2^(- 107 /128-1) + .quad 0x3FE3CCDD443B3394, 0x3FD1D4873168B9AA //2^( 108 /128-1) - 2^(- 108 /128-1), 2^(- 108 /128-1) + .quad 0x3FE4011B135B9590, 0x3FD1BBE084045CD4 //2^( 109 /128-1) - 2^(- 109 /128-1), 2^(- 109 /128-1) + .quad 0x3FE4357F544FA3C1, 0x3FD1A35BEB6FCB75 //2^( 110 /128-1) - 2^(- 110 /128-1), 2^(- 110 /128-1) + .quad 0x3FE46A0A6BC742FD, 0x3FD18AF9388C8DEA //2^( 111 /128-1) - 2^(- 111 /128-1), 2^(- 111 /128-1) + .quad 0x3FE49EBCBEBCFBCA, 0x3FD172B83C7D517B //2^( 112 /128-1) - 2^(- 112 /128-1), 2^(- 112 /128-1) + .quad 0x3FE4D396B276BC6F, 0x3FD15A98C8A58E51 //2^( 113 /128-1) - 2^(- 113 /128-1), 2^(- 113 /128-1) + .quad 0x3FE50898AC869B96, 0x3FD1429AAEA92DE0 //2^( 114 /128-1) - 2^(- 114 /128-1), 2^(- 114 /128-1) + .quad 0x3FE53DC312CB9B7A, 0x3FD12ABDC06C31CC //2^( 115 /128-1) - 2^(- 115 /128-1), 2^(- 115 /128-1) + .quad 0x3FE573164B726DB6, 0x3FD11301D0125B51 //2^( 116 /128-1) - 2^(- 116 /128-1), 2^(- 116 /128-1) + .quad 0x3FE5A892BCF6379B, 0x3FD0FB66AFFED31B //2^( 117 /128-1) - 2^(- 117 /128-1), 2^(- 117 /128-1) + .quad 0x3FE5DE38CE215725, 0x3FD0E3EC32D3D1A2 //2^( 118 /128-1) - 2^(- 118 /128-1), 2^(- 118 /128-1) + .quad 0x3FE61408E60E2888, 0x3FD0CC922B7247F7 //2^( 119 /128-1) - 2^(- 119 /128-1), 2^(- 119 /128-1) + .quad 0x3FE64A036C27CC52, 0x3FD0B5586CF9890F //2^( 120 /128-1) - 2^(- 120 /128-1), 2^(- 120 /128-1) + .quad 0x3FE68028C82AEE2F, 0x3FD09E3ECAC6F383 //2^( 121 /128-1) - 2^(- 121 /128-1), 2^(- 121 /128-1) + .quad 0x3FE6B67962268C43, 0x3FD0874518759BC8 //2^( 122 /128-1) - 2^(- 122 /128-1), 2^(- 122 /128-1) + .quad 0x3FE6ECF5A27CBF28, 0x3FD0706B29DDF6DE //2^( 123 /128-1) - 2^(- 123 /128-1), 2^(- 123 /128-1) + .quad 0x3FE7239DF1E38286, 0x3FD059B0D3158574 //2^( 124 /128-1) - 2^(- 124 /128-1), 2^(- 124 /128-1) + .quad 0x3FE75A72B9657E51, 0x3FD04315E86E7F85 //2^( 125 /128-1) - 2^(- 125 /128-1), 2^(- 125 /128-1) + .quad 0x3FE791746262D0A8, 0x3FD02C9A3E778061 //2^( 126 /128-1) - 2^(- 126 /128-1), 2^(- 126 /128-1) + .quad 0x3FE7C8A35691D856, 0x3FD0163DA9FB3335 //2^( 127 /128-1) - 2^(- 127 /128-1), 2^(- 127 /128-1) + .align 16 + .quad 0x42C8000000000000, 0x42C8000000000000 /* _dbShifter = 1.5 * 2^(52-k)*/ + .align 16 + .long 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99 /* _iDomainRange 0x40861d9ac12a3e85 =(1021*2^K-0.5)*log(2)/2^K -needed for quick exp*/ + .align 16 + .quad 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD /* _dPC2 */ + .align 16 + .quad 0x3FC55555555554AD, 0x3FC55555555554AD /* _dPC3 */ + .align 16 + .quad 0x3FA55555CF16D299, 0x3FA55555CF16D299 /* _dPC4 */ + .align 16 + .quad 0x3F8111115712F425, 0x3F8111115712F425 /* _dPC5 */ + .align 16 + .quad 0x000000000000007f, 0x000000000000007f /* _lIndexMask */ + .align 16 + .type __svml_dsinh_data_internal,@object + .size __svml_dsinh_data_internal,.-__svml_dsinh_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core-sse.S new file mode 100644 index 0000000000..ae531575fe --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core-sse.S @@ -0,0 +1,20 @@ +/* SSE version of vectorized sinh, 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 + . */ + +#define _ZGVdN4v_sinh _ZGVdN4v_sinh_sse_wrapper +#include "../svml_d_sinh4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core.c new file mode 100644 index 0000000000..bdf10b664b --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized sinh, 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 + . */ + +#define SYMBOL_NAME _ZGVdN4v_sinh +#include "ifunc-mathvec-avx2.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVdN4v_sinh, __GI__ZGVdN4v_sinh, __redirect__ZGVdN4v_sinh) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core_avx2.S new file mode 100644 index 0000000000..27b50d31a8 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh4_core_avx2.S @@ -0,0 +1,470 @@ +/* Function sinh 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_dsinh_data_internal + */ +#define _dbInvLn2 0 +#define _dbLn2hi 32 +#define _dbLn2lo 64 +#define _dSign 96 +#define _dbT 128 +#define _dbShifter 2176 +#define _iDomainRange 2208 +#define _dPC2 2240 +#define _dPC3 2272 +#define _dPC4 2304 +#define _dPC5 2336 +#define _lIndexMask 2368 + +#include + + .text + .section .text.avx2,"ax",@progbits +ENTRY(_ZGVdN4v_sinh_avx2) + pushq %rbp + cfi_def_cfa_offset(16) + movq %rsp, %rbp + cfi_def_cfa(6, 16) + cfi_offset(6, -16) + andq $-32, %rsp + subq $96, %rsp + lea _dbT+8+__svml_dsinh_data_internal(%rip), %r8 + vmovupd _dbShifter+__svml_dsinh_data_internal(%rip), %ymm12 + +/* + * Load argument + * dM = x*2^K/log(2) + RShifter + */ + vmovupd _dbInvLn2+__svml_dsinh_data_internal(%rip), %ymm5 + vmovupd _dbLn2hi+__svml_dsinh_data_internal(%rip), %ymm13 + vmovapd %ymm0, %ymm8 + +/* + * VLOAD_CONST( D, dPC[0], TAB._dPC1 ); + * Abs argument + */ + vandpd _dSign+__svml_dsinh_data_internal(%rip), %ymm8, %ymm7 + vxorpd %ymm8, %ymm7, %ymm6 + vfmadd213pd %ymm12, %ymm6, %ymm5 + +/* + * R + * dN = dM - RShifter + */ + vsubpd %ymm12, %ymm5, %ymm3 + +/* + * Index and lookup + * j + */ + vandps _lIndexMask+__svml_dsinh_data_internal(%rip), %ymm5, %ymm4 + +/* + * Check for overflow\underflow + * + */ + vextractf128 $1, %ymm6, %xmm9 + vshufps $221, %xmm9, %xmm6, %xmm10 + +/* dR = dX - dN*Log2_hi/2^K */ + vfnmadd231pd %ymm13, %ymm3, %ymm6 + vpcmpgtd _iDomainRange+__svml_dsinh_data_internal(%rip), %xmm10, %xmm11 + vmovmskps %xmm11, %eax + +/* dR = (dX - dN*Log2_hi/2^K) - dN*Log2_lo/2^K */ + vfnmadd231pd _dbLn2lo+__svml_dsinh_data_internal(%rip), %ymm3, %ymm6 + vextractf128 $1, %ymm4, %xmm0 + vmovd %xmm4, %edx + vmovd %xmm0, %esi + shll $4, %edx + vpextrd $2, %xmm4, %ecx + +/* split j and N */ + vxorps %ymm4, %ymm5, %ymm3 + shll $4, %esi + vpextrd $2, %xmm0, %edi + shll $4, %ecx + +/* + * G1,G2,G3: dTdif,dTn * 2^N,2^(-N) + * lM now is an EXP(2^N) + */ + vpsllq $45, %ymm3, %ymm4 + vmovq (%rdx,%r8), %xmm14 + vmovq (%rsi,%r8), %xmm1 + vmovhpd (%rcx,%r8), %xmm14, %xmm15 + shll $4, %edi + vmovhpd (%rdi,%r8), %xmm1, %xmm2 + +/* dR2 = dR^2 */ + vmulpd %ymm6, %ymm6, %ymm1 + vmovq -8(%rdx,%r8), %xmm9 + vmovq -8(%rsi,%r8), %xmm11 + vmovhpd -8(%rcx,%r8), %xmm9, %xmm10 + vmovhpd -8(%rdi,%r8), %xmm11, %xmm12 + vinsertf128 $1, %xmm2, %ymm15, %ymm2 + +/* */ + vpaddq %ymm4, %ymm2, %ymm5 + +/* */ + vpsubq %ymm4, %ymm2, %ymm14 + +/* dG3 = dTn*2^N + dTn*2^-N */ + vaddpd %ymm14, %ymm5, %ymm2 + +/* dG2 = dTn*2^N - dTn*2^-N */ + vsubpd %ymm14, %ymm5, %ymm14 + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*a5)) = r + r*(r^2*(a3+r^2*a5)) .... + * dSinh_r = (a3+r^2*a5) + */ + vmovupd _dPC5+__svml_dsinh_data_internal(%rip), %ymm5 + vfmadd213pd _dPC3+__svml_dsinh_data_internal(%rip), %ymm1, %ymm5 + vinsertf128 $1, %xmm12, %ymm10, %ymm13 + vpaddq %ymm4, %ymm13, %ymm0 + +/* dSinh_r = r^2*(a3+r^2*a5) */ + vmulpd %ymm5, %ymm1, %ymm4 + +/* dG2 += dG1 */ + vaddpd %ymm14, %ymm0, %ymm3 + +/* dG1 += dG3 */ + vaddpd %ymm2, %ymm0, %ymm0 + +/* dSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + vfmadd213pd %ymm6, %ymm6, %ymm4 + +/* + * poly(r) = (dG2+dG1)+dG3*sinh(dR)+dG1*sinh(dR)+(dG1+dG2)*dR2*(a2 +a4*dR2) + * dOut = (a2 +a4*dR2) + */ + vmovupd _dPC4+__svml_dsinh_data_internal(%rip), %ymm6 + vfmadd213pd _dPC2+__svml_dsinh_data_internal(%rip), %ymm1, %ymm6 + +/* dOut = dR2*(a2 +a4*dR2) */ + vmulpd %ymm6, %ymm1, %ymm1 + +/* dOut = dG2*dR2*(a2 +a4*dR2) */ + vmulpd %ymm3, %ymm1, %ymm6 + +/* dOut = dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + vfmadd213pd %ymm6, %ymm0, %ymm4 + +/* dOut = dG2 + dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + vaddpd %ymm4, %ymm3, %ymm5 + +/* Ret H */ + vorpd %ymm5, %ymm7, %ymm0 + testl %eax, %eax + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx r12 r13 r14 r15 eax 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 %ymm8, 32(%rsp) + vmovupd %ymm0, 64(%rsp) + # LOE rbx r12 r13 r14 r15 eax ymm0 + + xorl %edx, %edx + # 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: -80; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 + movl %edx, %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: -88; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 + movl %eax, %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: -96; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 64(%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: -80; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 + call sinh@PLT + # LOE rbx r14 r15 r12d r13d xmm0 + + movsd %xmm0, 64(%rsp,%r14,8) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx r15 r12d r13d +END(_ZGVdN4v_sinh_avx2) + + .section .rodata, "a" + .align 32 + +#ifdef __svml_dsinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(32)) VUINT32 _dbInvLn2[4][2]; + __declspec(align(32)) VUINT32 _dbLn2hi[4][2]; + __declspec(align(32)) VUINT32 _dbLn2lo[4][2]; + __declspec(align(32)) VUINT32 _dSign[4][2]; //0x8000000000000000 + __declspec(align(32)) VUINT32 _dbT[(1<<7)][2][2]; //precalc poly coeff + __declspec(align(32)) VUINT32 _dbShifter[4][2]; + __declspec(align(32)) VUINT32 _iDomainRange[8][1]; + __declspec(align(32)) VUINT32 _dPC2[4][2]; + __declspec(align(32)) VUINT32 _dPC3[4][2]; + __declspec(align(32)) VUINT32 _dPC4[4][2]; + __declspec(align(32)) VUINT32 _dPC5[4][2]; + __declspec(align(32)) VUINT32 _lIndexMask[4][2]; +} __svml_dsinh_data_internal; +#endif +__svml_dsinh_data_internal: + .quad 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE /* _dbInvLn2 = 1/log(2) */ + .align 32 + .quad 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000 /* _dbLn2hi = log(2) hi*/ + .align 32 + .quad 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A /* _dbLn2lo = log(2) lo*/ + .align 32 + .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 /* _dSign */ + //_dbT + .align 32 + .quad 0x0000000000000000, 0x3FE0000000000000 //2^( 0 /128-1) - 2^(- 0 /128-1), 2^(- 0 /128-1) + .quad 0x3F762E4A19BD1E74, 0x3FDFD3C22B8F71F1 //2^( 1 /128-1) - 2^(- 1 /128-1), 2^(- 1 /128-1) + .quad 0x3F862E5F6A0DFD36, 0x3FDFA7C1819E90D8 //2^( 2 /128-1) - 2^(- 2 /128-1), 2^(- 2 /128-1) + .quad 0x3F90A2E234040F5F, 0x3FDF7BFDAD9CBE14 //2^( 3 /128-1) - 2^(- 3 /128-1), 2^(- 3 /128-1) + .quad 0x3F962EB4ABCC5A81, 0x3FDF50765B6E4540 //2^( 4 /128-1) - 2^(- 4 /128-1), 2^(- 4 /128-1) + .quad 0x3F9BBAB1C5033244, 0x3FDF252B376BBA97 //2^( 5 /128-1) - 2^(- 5 /128-1), 2^(- 5 /128-1) + .quad 0x3FA0A372144EEB45, 0x3FDEFA1BEE615A27 //2^( 6 /128-1) - 2^(- 6 /128-1), 2^(- 6 /128-1) + .quad 0x3FA369AB3FFBF8B0, 0x3FDECF482D8E67F1 //2^( 7 /128-1) - 2^(- 7 /128-1), 2^(- 7 /128-1) + .quad 0x3FA63009BA740A2A, 0x3FDEA4AFA2A490DA //2^( 8 /128-1) - 2^(- 8 /128-1), 2^(- 8 /128-1) + .quad 0x3FA8F692D8EA1B5A, 0x3FDE7A51FBC74C83 //2^( 9 /128-1) - 2^(- 9 /128-1), 2^(- 9 /128-1) + .quad 0x3FABBD4BF0E31A6F, 0x3FDE502EE78B3FF6 //2^( 10 /128-1) - 2^(- 10 /128-1), 2^(- 10 /128-1) + .quad 0x3FAE843A5840286A, 0x3FDE264614F5A129 //2^( 11 /128-1) - 2^(- 11 /128-1), 2^(- 11 /128-1) + .quad 0x3FB0A5B1B2A46D0A, 0x3FDDFC97337B9B5F //2^( 12 /128-1) - 2^(- 12 /128-1), 2^(- 12 /128-1) + .quad 0x3FB20966375ABCDF, 0x3FDDD321F301B460 //2^( 13 /128-1) - 2^(- 13 /128-1), 2^(- 13 /128-1) + .quad 0x3FB36D3D65DCA4E8, 0x3FDDA9E603DB3285 //2^( 14 /128-1) - 2^(- 14 /128-1), 2^(- 14 /128-1) + .quad 0x3FB4D139EA06642A, 0x3FDD80E316C98398 //2^( 15 /128-1) - 2^(- 15 /128-1), 2^(- 15 /128-1) + .quad 0x3FB6355E6FFBF9BA, 0x3FDD5818DCFBA487 //2^( 16 /128-1) - 2^(- 16 /128-1), 2^(- 16 /128-1) + .quad 0x3FB799ADA42E4788, 0x3FDD2F87080D89F2 //2^( 17 /128-1) - 2^(- 17 /128-1), 2^(- 17 /128-1) + .quad 0x3FB8FE2A336035BC, 0x3FDD072D4A07897C //2^( 18 /128-1) - 2^(- 18 /128-1), 2^(- 18 /128-1) + .quad 0x3FBA62D6CAABD6B6, 0x3FDCDF0B555DC3FA //2^( 19 /128-1) - 2^(- 19 /128-1), 2^(- 19 /128-1) + .quad 0x3FBBC7B617878BAF, 0x3FDCB720DCEF9069 //2^( 20 /128-1) - 2^(- 20 /128-1), 2^(- 20 /128-1) + .quad 0x3FBD2CCAC7CB2A11, 0x3FDC8F6D9406E7B5 //2^( 21 /128-1) - 2^(- 21 /128-1), 2^(- 21 /128-1) + .quad 0x3FBE921789B52185, 0x3FDC67F12E57D14B //2^( 22 /128-1) - 2^(- 22 /128-1), 2^(- 22 /128-1) + .quad 0x3FBFF79F0BEFA2C7, 0x3FDC40AB5FFFD07A //2^( 23 /128-1) - 2^(- 23 /128-1), 2^(- 23 /128-1) + .quad 0x3FC0AEB1FECAE3A9, 0x3FDC199BDD85529C //2^( 24 /128-1) - 2^(- 24 /128-1), 2^(- 24 /128-1) + .quad 0x3FC161B4871C5CEC, 0x3FDBF2C25BD71E09 //2^( 25 /128-1) - 2^(- 25 /128-1), 2^(- 25 /128-1) + .quad 0x3FC214D876F26FD0, 0x3FDBCC1E904BC1D2 //2^( 26 /128-1) - 2^(- 26 /128-1), 2^(- 26 /128-1) + .quad 0x3FC2C81F2693816F, 0x3FDBA5B030A1064A //2^( 27 /128-1) - 2^(- 27 /128-1), 2^(- 27 /128-1) + .quad 0x3FC37B89EE88BEF7, 0x3FDB7F76F2FB5E47 //2^( 28 /128-1) - 2^(- 28 /128-1), 2^(- 28 /128-1) + .quad 0x3FC42F1A27A0B3CD, 0x3FDB59728DE5593A //2^( 29 /128-1) - 2^(- 29 /128-1), 2^(- 29 /128-1) + .quad 0x3FC4E2D12AF1E037, 0x3FDB33A2B84F15FB //2^( 30 /128-1) - 2^(- 30 /128-1), 2^(- 30 /128-1) + .quad 0x3FC596B051DD508D, 0x3FDB0E07298DB666 //2^( 31 /128-1) - 2^(- 31 /128-1), 2^(- 31 /128-1) + .quad 0x3FC64AB8F61134FA, 0x3FDAE89F995AD3AD //2^( 32 /128-1) - 2^(- 32 /128-1), 2^(- 32 /128-1) + .quad 0x3FC6FEEC718B79D1, 0x3FDAC36BBFD3F37A //2^( 33 /128-1) - 2^(- 33 /128-1), 2^(- 33 /128-1) + .quad 0x3FC7B34C1E9C607F, 0x3FDA9E6B5579FDBF //2^( 34 /128-1) - 2^(- 34 /128-1), 2^(- 34 /128-1) + .quad 0x3FC867D957E91912, 0x3FDA799E1330B358 //2^( 35 /128-1) - 2^(- 35 /128-1), 2^(- 35 /128-1) + .quad 0x3FC91C95786E5C72, 0x3FDA5503B23E255D //2^( 36 /128-1) - 2^(- 36 /128-1), 2^(- 36 /128-1) + .quad 0x3FC9D181DB83072F, 0x3FDA309BEC4A2D33 //2^( 37 /128-1) - 2^(- 37 /128-1), 2^(- 37 /128-1) + .quad 0x3FCA869FDCDAB512, 0x3FDA0C667B5DE565 //2^( 38 /128-1) - 2^(- 38 /128-1), 2^(- 38 /128-1) + .quad 0x3FCB3BF0D8885D4C, 0x3FD9E86319E32323 //2^( 39 /128-1) - 2^(- 39 /128-1), 2^(- 39 /128-1) + .quad 0x3FCBF1762B00EF69, 0x3FD9C49182A3F090 //2^( 40 /128-1) - 2^(- 40 /128-1), 2^(- 40 /128-1) + .quad 0x3FCCA731311DF0FB, 0x3FD9A0F170CA07BA //2^( 41 /128-1) - 2^(- 41 /128-1), 2^(- 41 /128-1) + .quad 0x3FCD5D2348201C09, 0x3FD97D829FDE4E50 //2^( 42 /128-1) - 2^(- 42 /128-1), 2^(- 42 /128-1) + .quad 0x3FCE134DCDB1FE3E, 0x3FD95A44CBC8520F //2^( 43 /128-1) - 2^(- 43 /128-1), 2^(- 43 /128-1) + .quad 0x3FCEC9B21FEA98EA, 0x3FD93737B0CDC5E5 //2^( 44 /128-1) - 2^(- 44 /128-1), 2^(- 44 /128-1) + .quad 0x3FCF80519D5001D3, 0x3FD9145B0B91FFC6 //2^( 45 /128-1) - 2^(- 45 /128-1), 2^(- 45 /128-1) + .quad 0x3FD01B96D26D026A, 0x3FD8F1AE99157736 //2^( 46 /128-1) - 2^(- 46 /128-1), 2^(- 46 /128-1) + .quad 0x3FD07723CAFA6331, 0x3FD8CF3216B5448C //2^( 47 /128-1) - 2^(- 47 /128-1), 2^(- 47 /128-1) + .quad 0x3FD0D2D06841B373, 0x3FD8ACE5422AA0DB //2^( 48 /128-1) - 2^(- 48 /128-1), 2^(- 48 /128-1) + .quad 0x3FD12E9D5A715381, 0x3FD88AC7D98A6699 //2^( 49 /128-1) - 2^(- 49 /128-1), 2^(- 49 /128-1) + .quad 0x3FD18A8B51F5C661, 0x3FD868D99B4492ED //2^( 50 /128-1) - 2^(- 50 /128-1), 2^(- 50 /128-1) + .quad 0x3FD1E69AFF7B04D7, 0x3FD8471A4623C7AD //2^( 51 /128-1) - 2^(- 51 /128-1), 2^(- 51 /128-1) + .quad 0x3FD242CD13EDD0F1, 0x3FD82589994CCE13 //2^( 52 /128-1) - 2^(- 52 /128-1), 2^(- 52 /128-1) + .quad 0x3FD29F22407D0A0C, 0x3FD80427543E1A12 //2^( 53 /128-1) - 2^(- 53 /128-1), 2^(- 53 /128-1) + .quad 0x3FD2FB9B369B0153, 0x3FD7E2F336CF4E62 //2^( 54 /128-1) - 2^(- 54 /128-1), 2^(- 54 /128-1) + .quad 0x3FD35838A7FECEC8, 0x3FD7C1ED0130C132 //2^( 55 /128-1) - 2^(- 55 /128-1), 2^(- 55 /128-1) + .quad 0x3FD3B4FB46A5A6CC, 0x3FD7A11473EB0187 //2^( 56 /128-1) - 2^(- 56 /128-1), 2^(- 56 /128-1) + .quad 0x3FD411E3C4D4302F, 0x3FD780694FDE5D3F //2^( 57 /128-1) - 2^(- 57 /128-1), 2^(- 57 /128-1) + .quad 0x3FD46EF2D517DAC8, 0x3FD75FEB564267C9 //2^( 58 /128-1) - 2^(- 58 /128-1), 2^(- 58 /128-1) + .quad 0x3FD4CC292A48369E, 0x3FD73F9A48A58174 //2^( 59 /128-1) - 2^(- 59 /128-1), 2^(- 59 /128-1) + .quad 0x3FD5298777884B96, 0x3FD71F75E8EC5F74 //2^( 60 /128-1) - 2^(- 60 /128-1), 2^(- 60 /128-1) + .quad 0x3FD5870E7047F1BC, 0x3FD6FF7DF9519484 //2^( 61 /128-1) - 2^(- 61 /128-1), 2^(- 61 /128-1) + .quad 0x3FD5E4BEC8452A1A, 0x3FD6DFB23C651A2F //2^( 62 /128-1) - 2^(- 62 /128-1), 2^(- 62 /128-1) + .quad 0x3FD64299338D7827, 0x3FD6C012750BDABF //2^( 63 /128-1) - 2^(- 63 /128-1), 2^(- 63 /128-1) + .quad 0x3FD6A09E667F3BCD, 0x3FD6A09E667F3BCD //2^( 64 /128-1) - 2^(- 64 /128-1), 2^(- 64 /128-1) + .quad 0x3FD6FECF15CB0C0B, 0x3FD68155D44CA973 //2^( 65 /128-1) - 2^(- 65 /128-1), 2^(- 65 /128-1) + .quad 0x3FD75D2BF6751239, 0x3FD6623882552225 //2^( 66 /128-1) - 2^(- 66 /128-1), 2^(- 66 /128-1) + .quad 0x3FD7BBB5BDD665E8, 0x3FD6434634CCC320 //2^( 67 /128-1) - 2^(- 67 /128-1), 2^(- 67 /128-1) + .quad 0x3FD81A6D219E6963, 0x3FD6247EB03A5585 //2^( 68 /128-1) - 2^(- 68 /128-1), 2^(- 68 /128-1) + .quad 0x3FD87952D7D426DF, 0x3FD605E1B976DC09 //2^( 69 /128-1) - 2^(- 69 /128-1), 2^(- 69 /128-1) + .quad 0x3FD8D86796D7AE49, 0x3FD5E76F15AD2148 //2^( 70 /128-1) - 2^(- 70 /128-1), 2^(- 70 /128-1) + .quad 0x3FD937AC156373C8, 0x3FD5C9268A5946B7 //2^( 71 /128-1) - 2^(- 71 /128-1), 2^(- 71 /128-1) + .quad 0x3FD997210A8DAEE4, 0x3FD5AB07DD485429 //2^( 72 /128-1) - 2^(- 72 /128-1), 2^(- 72 /128-1) + .quad 0x3FD9F6C72DC9BA68, 0x3FD58D12D497C7FD //2^( 73 /128-1) - 2^(- 73 /128-1), 2^(- 73 /128-1) + .quad 0x3FDA569F36E974EA, 0x3FD56F4736B527DA //2^( 74 /128-1) - 2^(- 74 /128-1), 2^(- 74 /128-1) + .quad 0x3FDAB6A9DE1EA215, 0x3FD551A4CA5D920F //2^( 75 /128-1) - 2^(- 75 /128-1), 2^(- 75 /128-1) + .quad 0x3FDB16E7DBFC4CA3, 0x3FD5342B569D4F82 //2^( 76 /128-1) - 2^(- 76 /128-1), 2^(- 76 /128-1) + .quad 0x3FDB7759E9782918, 0x3FD516DAA2CF6642 //2^( 77 /128-1) - 2^(- 77 /128-1), 2^(- 77 /128-1) + .quad 0x3FDBD800BFEBF932, 0x3FD4F9B2769D2CA7 //2^( 78 /128-1) - 2^(- 78 /128-1), 2^(- 78 /128-1) + .quad 0x3FDC38DD1916F025, 0x3FD4DCB299FDDD0D //2^( 79 /128-1) - 2^(- 79 /128-1), 2^(- 79 /128-1) + .quad 0x3FDC99EFAF1F1790, 0x3FD4BFDAD5362A27 //2^( 80 /128-1) - 2^(- 80 /128-1), 2^(- 80 /128-1) + .quad 0x3FDCFB393C92B539, 0x3FD4A32AF0D7D3DE //2^( 81 /128-1) - 2^(- 81 /128-1), 2^(- 81 /128-1) + .quad 0x3FDD5CBA7C69B19C, 0x3FD486A2B5C13CD0 //2^( 82 /128-1) - 2^(- 82 /128-1), 2^(- 82 /128-1) + .quad 0x3FDDBE742A06FF34, 0x3FD46A41ED1D0057 //2^( 83 /128-1) - 2^(- 83 /128-1), 2^(- 83 /128-1) + .quad 0x3FDE2067013A029D, 0x3FD44E086061892D //2^( 84 /128-1) - 2^(- 84 /128-1), 2^(- 84 /128-1) + .quad 0x3FDE8293BE3FFB87, 0x3FD431F5D950A897 //2^( 85 /128-1) - 2^(- 85 /128-1), 2^(- 85 /128-1) + .quad 0x3FDEE4FB1DC56E75, 0x3FD4160A21F72E2A //2^( 86 /128-1) - 2^(- 86 /128-1), 2^(- 86 /128-1) + .quad 0x3FDF479DDCE78F58, 0x3FD3FA4504AC801C //2^( 87 /128-1) - 2^(- 87 /128-1), 2^(- 87 /128-1) + .quad 0x3FDFAA7CB935ACFE, 0x3FD3DEA64C123422 //2^( 88 /128-1) - 2^(- 88 /128-1), 2^(- 88 /128-1) + .quad 0x3FE006CC38594EB1, 0x3FD3C32DC313A8E5 //2^( 89 /128-1) - 2^(- 89 /128-1), 2^(- 89 /128-1) + .quad 0x3FE03878E0EB1569, 0x3FD3A7DB34E59FF7 //2^( 90 /128-1) - 2^(- 90 /128-1), 2^(- 90 /128-1) + .quad 0x3FE06A44B5C74101, 0x3FD38CAE6D05D866 //2^( 91 /128-1) - 2^(- 91 /128-1), 2^(- 91 /128-1) + .quad 0x3FE09C3016A0D077, 0x3FD371A7373AA9CB //2^( 92 /128-1) - 2^(- 92 /128-1), 2^(- 92 /128-1) + .quad 0x3FE0CE3B63676360, 0x3FD356C55F929FF1 //2^( 93 /128-1) - 2^(- 93 /128-1), 2^(- 93 /128-1) + .quad 0x3FE10066FC47F240, 0x3FD33C08B26416FF //2^( 94 /128-1) - 2^(- 94 /128-1), 2^(- 94 /128-1) + .quad 0x3FE132B341AD8761, 0x3FD32170FC4CD831 //2^( 95 /128-1) - 2^(- 95 /128-1), 2^(- 95 /128-1) + .quad 0x3FE165209441F823, 0x3FD306FE0A31B715 //2^( 96 /128-1) - 2^(- 96 /128-1), 2^(- 96 /128-1) + .quad 0x3FE197AF54EE9EBB, 0x3FD2ECAFA93E2F56 //2^( 97 /128-1) - 2^(- 97 /128-1), 2^(- 97 /128-1) + .quad 0x3FE1CA5FE4DD1475, 0x3FD2D285A6E4030B //2^( 98 /128-1) - 2^(- 98 /128-1), 2^(- 98 /128-1) + .quad 0x3FE1FD32A577EC72, 0x3FD2B87FD0DAD990 //2^( 99 /128-1) - 2^(- 99 /128-1), 2^(- 99 /128-1) + .quad 0x3FE23027F86B6ED6, 0x3FD29E9DF51FDEE1 //2^( 100 /128-1) - 2^(- 100 /128-1), 2^(- 100 /128-1) + .quad 0x3FE263403FA65489, 0x3FD284DFE1F56381 //2^( 101 /128-1) - 2^(- 101 /128-1), 2^(- 101 /128-1) + .quad 0x3FE2967BDD5A8364, 0x3FD26B4565E27CDD //2^( 102 /128-1) - 2^(- 102 /128-1), 2^(- 102 /128-1) + .quad 0x3FE2C9DB33FDCAE9, 0x3FD251CE4FB2A63F //2^( 103 /128-1) - 2^(- 103 /128-1), 2^(- 103 /128-1) + .quad 0x3FE2FD5EA64AA180, 0x3FD2387A6E756238 //2^( 104 /128-1) - 2^(- 104 /128-1), 2^(- 104 /128-1) + .quad 0x3FE331069740E22F, 0x3FD21F49917DDC96 //2^( 105 /128-1) - 2^(- 105 /128-1), 2^(- 105 /128-1) + .quad 0x3FE364D36A268AE0, 0x3FD2063B88628CD6 //2^( 106 /128-1) - 2^(- 106 /128-1), 2^(- 106 /128-1) + .quad 0x3FE398C582887B27, 0x3FD1ED5022FCD91D //2^( 107 /128-1) - 2^(- 107 /128-1), 2^(- 107 /128-1) + .quad 0x3FE3CCDD443B3394, 0x3FD1D4873168B9AA //2^( 108 /128-1) - 2^(- 108 /128-1), 2^(- 108 /128-1) + .quad 0x3FE4011B135B9590, 0x3FD1BBE084045CD4 //2^( 109 /128-1) - 2^(- 109 /128-1), 2^(- 109 /128-1) + .quad 0x3FE4357F544FA3C1, 0x3FD1A35BEB6FCB75 //2^( 110 /128-1) - 2^(- 110 /128-1), 2^(- 110 /128-1) + .quad 0x3FE46A0A6BC742FD, 0x3FD18AF9388C8DEA //2^( 111 /128-1) - 2^(- 111 /128-1), 2^(- 111 /128-1) + .quad 0x3FE49EBCBEBCFBCA, 0x3FD172B83C7D517B //2^( 112 /128-1) - 2^(- 112 /128-1), 2^(- 112 /128-1) + .quad 0x3FE4D396B276BC6F, 0x3FD15A98C8A58E51 //2^( 113 /128-1) - 2^(- 113 /128-1), 2^(- 113 /128-1) + .quad 0x3FE50898AC869B96, 0x3FD1429AAEA92DE0 //2^( 114 /128-1) - 2^(- 114 /128-1), 2^(- 114 /128-1) + .quad 0x3FE53DC312CB9B7A, 0x3FD12ABDC06C31CC //2^( 115 /128-1) - 2^(- 115 /128-1), 2^(- 115 /128-1) + .quad 0x3FE573164B726DB6, 0x3FD11301D0125B51 //2^( 116 /128-1) - 2^(- 116 /128-1), 2^(- 116 /128-1) + .quad 0x3FE5A892BCF6379B, 0x3FD0FB66AFFED31B //2^( 117 /128-1) - 2^(- 117 /128-1), 2^(- 117 /128-1) + .quad 0x3FE5DE38CE215725, 0x3FD0E3EC32D3D1A2 //2^( 118 /128-1) - 2^(- 118 /128-1), 2^(- 118 /128-1) + .quad 0x3FE61408E60E2888, 0x3FD0CC922B7247F7 //2^( 119 /128-1) - 2^(- 119 /128-1), 2^(- 119 /128-1) + .quad 0x3FE64A036C27CC52, 0x3FD0B5586CF9890F //2^( 120 /128-1) - 2^(- 120 /128-1), 2^(- 120 /128-1) + .quad 0x3FE68028C82AEE2F, 0x3FD09E3ECAC6F383 //2^( 121 /128-1) - 2^(- 121 /128-1), 2^(- 121 /128-1) + .quad 0x3FE6B67962268C43, 0x3FD0874518759BC8 //2^( 122 /128-1) - 2^(- 122 /128-1), 2^(- 122 /128-1) + .quad 0x3FE6ECF5A27CBF28, 0x3FD0706B29DDF6DE //2^( 123 /128-1) - 2^(- 123 /128-1), 2^(- 123 /128-1) + .quad 0x3FE7239DF1E38286, 0x3FD059B0D3158574 //2^( 124 /128-1) - 2^(- 124 /128-1), 2^(- 124 /128-1) + .quad 0x3FE75A72B9657E51, 0x3FD04315E86E7F85 //2^( 125 /128-1) - 2^(- 125 /128-1), 2^(- 125 /128-1) + .quad 0x3FE791746262D0A8, 0x3FD02C9A3E778061 //2^( 126 /128-1) - 2^(- 126 /128-1), 2^(- 126 /128-1) + .quad 0x3FE7C8A35691D856, 0x3FD0163DA9FB3335 //2^( 127 /128-1) - 2^(- 127 /128-1), 2^(- 127 /128-1) + .align 32 + .quad 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000 /* _dbShifter = 1.5 * 2^(52-k)*/ + .align 32 + .long 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99 /* _iDomainRange 0x40861d9ac12a3e85 =(1021*2^K-0.5)*log(2)/2^K -needed for quick exp*/ + .align 32 + .quad 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD /* _dPC2 */ + .align 32 + .quad 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD /* _dPC3 */ + .align 32 + .quad 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299 /* _dPC4 */ + .align 32 + .quad 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425 /* _dPC5 */ + .align 32 + .quad 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f /* _lIndexMask */ + .align 32 + .type __svml_dsinh_data_internal,@object + .size __svml_dsinh_data_internal,.-__svml_dsinh_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core-avx2.S new file mode 100644 index 0000000000..d767d25080 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core-avx2.S @@ -0,0 +1,20 @@ +/* AVX2 version of vectorized sinh, 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 + . */ + +#define _ZGVeN8v_sinh _ZGVeN8v_sinh_avx2_wrapper +#include "../svml_d_sinh8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core.c new file mode 100644 index 0000000000..427d07bce2 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized sinh, 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 + . */ + +#define SYMBOL_NAME _ZGVeN8v_sinh +#include "ifunc-mathvec-avx512-skx.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVeN8v_sinh, __GI__ZGVeN8v_sinh, __redirect__ZGVeN8v_sinh) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core_avx512.S new file mode 100644 index 0000000000..d057d6c7eb --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sinh8_core_avx512.S @@ -0,0 +1,461 @@ +/* Function sinh 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_dsinh_data_internal + */ +#define _dbInvLn2 0 +#define _dbLn2hi 64 +#define _dbLn2lo 128 +#define _dSign 192 +#define _dbT 256 +#define _dbShifter 2304 +#define _iDomainRange 2368 +#define _dPC2 2432 +#define _dPC3 2496 +#define _dPC4 2560 +#define _dPC5 2624 +#define _lIndexMask 2688 + +#include + + .text + .section .text.evex512,"ax",@progbits +ENTRY(_ZGVeN8v_sinh_skx) + pushq %rbp + cfi_def_cfa_offset(16) + movq %rsp, %rbp + cfi_def_cfa(6, 16) + cfi_offset(6, -16) + andq $-64, %rsp + subq $192, %rsp + lea _dbT+8+__svml_dsinh_data_internal(%rip), %rax + vmovaps %zmm0, %zmm8 + +/* Abs argument */ + vandpd _dSign+__svml_dsinh_data_internal(%rip), %zmm8, %zmm7 + vmovups _dbShifter+__svml_dsinh_data_internal(%rip), %zmm13 + +/* + * Load argument + * dM = x*2^K/log(2) + RShifter + */ + vmovups _dbInvLn2+__svml_dsinh_data_internal(%rip), %zmm12 + vmovups _dbLn2hi+__svml_dsinh_data_internal(%rip), %zmm14 + vmovups _dPC5+__svml_dsinh_data_internal(%rip), %zmm6 + +/* VLOAD_CONST( D, dPC[0], TAB._dPC1 ); */ + vmovups _dPC4+__svml_dsinh_data_internal(%rip), %zmm4 + vxorpd %zmm8, %zmm7, %zmm5 + kxnorw %k0, %k0, %k1 + kxnorw %k0, %k0, %k2 + vfmadd213pd {rn-sae}, %zmm13, %zmm5, %zmm12 + +/* + * Check for overflow\underflow + * + */ + vpsrlq $32, %zmm5, %zmm9 + +/* + * R + * dN = dM - RShifter + */ + vsubpd {rn-sae}, %zmm13, %zmm12, %zmm2 + vpmovqd %zmm9, %ymm10 + vmovups _dbLn2lo+__svml_dsinh_data_internal(%rip), %zmm9 + +/* dR = dX - dN*Log2_hi/2^K */ + vfnmadd231pd {rn-sae}, %zmm14, %zmm2, %zmm5 + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*a5)) = r + r*(r^2*(a3+r^2*a5)) .... + * dSinh_r = (a3+r^2*a5) + */ + vmovups _dPC3+__svml_dsinh_data_internal(%rip), %zmm14 + +/* dR = (dX - dN*Log2_hi/2^K) - dN*Log2_lo/2^K */ + vfnmadd231pd {rn-sae}, %zmm9, %zmm2, %zmm5 + vpcmpgtd _iDomainRange+__svml_dsinh_data_internal(%rip), %ymm10, %ymm11 + vmovmskps %ymm11, %edx + +/* dR2 = dR^2 */ + vmulpd {rn-sae}, %zmm5, %zmm5, %zmm2 + vfmadd231pd {rn-sae}, %zmm2, %zmm6, %zmm14 + +/* + * Index and lookup + * j + */ + vpandq _lIndexMask+__svml_dsinh_data_internal(%rip), %zmm12, %zmm15 + vpsllq $4, %zmm15, %zmm1 + vpmovqd %zmm1, %ymm0 + vpxord %zmm11, %zmm11, %zmm11 + vpxord %zmm10, %zmm10, %zmm10 + vgatherdpd (%rax,%ymm0), %zmm11{%k1} + vgatherdpd -8(%rax,%ymm0), %zmm10{%k2} + +/* split j and N */ + vpxorq %zmm15, %zmm12, %zmm3 + +/* + * G1,G2,G3: dTdif,dTn * 2^N,2^(-N) + * lM now is an EXP(2^N) + */ + vpsllq $45, %zmm3, %zmm3 + vpaddq %zmm3, %zmm10, %zmm1 + +/* */ + vpaddq %zmm3, %zmm11, %zmm12 + +/* */ + vpsubq %zmm3, %zmm11, %zmm13 + +/* dSinh_r = r^2*(a3+r^2*a5) */ + vmulpd {rn-sae}, %zmm2, %zmm14, %zmm3 + +/* dG2 = dTn*2^N - dTn*2^-N */ + vsubpd {rn-sae}, %zmm13, %zmm12, %zmm15 + +/* dG3 = dTn*2^N + dTn*2^-N */ + vaddpd {rn-sae}, %zmm13, %zmm12, %zmm0 + +/* dSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + vfmadd213pd {rn-sae}, %zmm5, %zmm5, %zmm3 + +/* + * poly(r) = (dG2+dG1)+dG3*sinh(dR)+dG1*sinh(dR)+(dG1+dG2)*dR2*(a2 +a4*dR2) + * dOut = (a2 +a4*dR2) + */ + vmovups _dPC2+__svml_dsinh_data_internal(%rip), %zmm5 + +/* dG1 += dG3 */ + vaddpd {rn-sae}, %zmm0, %zmm1, %zmm6 + vfmadd231pd {rn-sae}, %zmm2, %zmm4, %zmm5 + +/* dOut = dR2*(a2 +a4*dR2) */ + vmulpd {rn-sae}, %zmm2, %zmm5, %zmm4 + +/* dG2 += dG1 */ + vaddpd {rn-sae}, %zmm15, %zmm1, %zmm2 + +/* dOut = dG2*dR2*(a2 +a4*dR2) */ + vmulpd {rn-sae}, %zmm2, %zmm4, %zmm4 + +/* dOut = dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + vfmadd213pd {rn-sae}, %zmm4, %zmm6, %zmm3 + +/* dOut = dG2 + dG1*sinh(dR)+dG2*dR2*(a2 +a4*dR2) */ + vaddpd {rn-sae}, %zmm2, %zmm3, %zmm0 + +/* Ret H */ + vorpd %zmm0, %zmm7, %zmm0 + testl %edx, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm8 + +/* 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): + vmovups %zmm8, 64(%rsp) + vmovups %zmm0, 128(%rsp) + # LOE rbx r12 r13 r14 r15 edx zmm0 + + 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: -176; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 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 128(%rsp), %zmm0 + +/* 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: -176; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 + # LOE rbx r12 r13 r14 r15 zmm0 + +/* Scalar math fucntion call + * to process special input + */ + +L(SCALAR_MATH_CALL): + movl %r12d, %r14d + movsd 64(%rsp,%r14,8), %xmm0 + call sinh@PLT + # LOE rbx r14 r15 r12d r13d xmm0 + + movsd %xmm0, 128(%rsp,%r14,8) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx r15 r12d r13d +END(_ZGVeN8v_sinh_skx) + + .section .rodata, "a" + .align 64 + +#ifdef __svml_dsinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(64)) VUINT32 _dbInvLn2[8][2]; + __declspec(align(64)) VUINT32 _dbLn2hi[8][2]; + __declspec(align(64)) VUINT32 _dbLn2lo[8][2]; + __declspec(align(64)) VUINT32 _dSign[8][2]; //0x8000000000000000 + __declspec(align(64)) VUINT32 _dbT[(1<<7)][2][2]; //precalc poly coeff + __declspec(align(64)) VUINT32 _dbShifter[8][2]; + __declspec(align(64)) VUINT32 _iDomainRange[16][1]; + __declspec(align(64)) VUINT32 _dPC2[8][2]; + __declspec(align(64)) VUINT32 _dPC3[8][2]; + __declspec(align(64)) VUINT32 _dPC4[8][2]; + __declspec(align(64)) VUINT32 _dPC5[8][2]; + __declspec(align(64)) VUINT32 _lIndexMask[8][2]; +} __svml_dsinh_data_internal; +#endif +__svml_dsinh_data_internal: + .quad 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE, 0x3FF71547652B82FE /* _dbInvLn2 = 1/log(2) */ + .align 64 + .quad 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000, 0x3FE62E42FEFA0000 /* _dbLn2hi = log(2) hi*/ + .align 64 + .quad 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A, 0x3D7CF79ABC9E3B3A /* _dbLn2lo = log(2) lo*/ + .align 64 + .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 /* _dSign */ + //_dbT + .align 64 + .quad 0x0000000000000000, 0x3FE0000000000000 //2^( 0 /128-1) - 2^(- 0 /128-1), 2^(- 0 /128-1) + .quad 0x3F762E4A19BD1E74, 0x3FDFD3C22B8F71F1 //2^( 1 /128-1) - 2^(- 1 /128-1), 2^(- 1 /128-1) + .quad 0x3F862E5F6A0DFD36, 0x3FDFA7C1819E90D8 //2^( 2 /128-1) - 2^(- 2 /128-1), 2^(- 2 /128-1) + .quad 0x3F90A2E234040F5F, 0x3FDF7BFDAD9CBE14 //2^( 3 /128-1) - 2^(- 3 /128-1), 2^(- 3 /128-1) + .quad 0x3F962EB4ABCC5A81, 0x3FDF50765B6E4540 //2^( 4 /128-1) - 2^(- 4 /128-1), 2^(- 4 /128-1) + .quad 0x3F9BBAB1C5033244, 0x3FDF252B376BBA97 //2^( 5 /128-1) - 2^(- 5 /128-1), 2^(- 5 /128-1) + .quad 0x3FA0A372144EEB45, 0x3FDEFA1BEE615A27 //2^( 6 /128-1) - 2^(- 6 /128-1), 2^(- 6 /128-1) + .quad 0x3FA369AB3FFBF8B0, 0x3FDECF482D8E67F1 //2^( 7 /128-1) - 2^(- 7 /128-1), 2^(- 7 /128-1) + .quad 0x3FA63009BA740A2A, 0x3FDEA4AFA2A490DA //2^( 8 /128-1) - 2^(- 8 /128-1), 2^(- 8 /128-1) + .quad 0x3FA8F692D8EA1B5A, 0x3FDE7A51FBC74C83 //2^( 9 /128-1) - 2^(- 9 /128-1), 2^(- 9 /128-1) + .quad 0x3FABBD4BF0E31A6F, 0x3FDE502EE78B3FF6 //2^( 10 /128-1) - 2^(- 10 /128-1), 2^(- 10 /128-1) + .quad 0x3FAE843A5840286A, 0x3FDE264614F5A129 //2^( 11 /128-1) - 2^(- 11 /128-1), 2^(- 11 /128-1) + .quad 0x3FB0A5B1B2A46D0A, 0x3FDDFC97337B9B5F //2^( 12 /128-1) - 2^(- 12 /128-1), 2^(- 12 /128-1) + .quad 0x3FB20966375ABCDF, 0x3FDDD321F301B460 //2^( 13 /128-1) - 2^(- 13 /128-1), 2^(- 13 /128-1) + .quad 0x3FB36D3D65DCA4E8, 0x3FDDA9E603DB3285 //2^( 14 /128-1) - 2^(- 14 /128-1), 2^(- 14 /128-1) + .quad 0x3FB4D139EA06642A, 0x3FDD80E316C98398 //2^( 15 /128-1) - 2^(- 15 /128-1), 2^(- 15 /128-1) + .quad 0x3FB6355E6FFBF9BA, 0x3FDD5818DCFBA487 //2^( 16 /128-1) - 2^(- 16 /128-1), 2^(- 16 /128-1) + .quad 0x3FB799ADA42E4788, 0x3FDD2F87080D89F2 //2^( 17 /128-1) - 2^(- 17 /128-1), 2^(- 17 /128-1) + .quad 0x3FB8FE2A336035BC, 0x3FDD072D4A07897C //2^( 18 /128-1) - 2^(- 18 /128-1), 2^(- 18 /128-1) + .quad 0x3FBA62D6CAABD6B6, 0x3FDCDF0B555DC3FA //2^( 19 /128-1) - 2^(- 19 /128-1), 2^(- 19 /128-1) + .quad 0x3FBBC7B617878BAF, 0x3FDCB720DCEF9069 //2^( 20 /128-1) - 2^(- 20 /128-1), 2^(- 20 /128-1) + .quad 0x3FBD2CCAC7CB2A11, 0x3FDC8F6D9406E7B5 //2^( 21 /128-1) - 2^(- 21 /128-1), 2^(- 21 /128-1) + .quad 0x3FBE921789B52185, 0x3FDC67F12E57D14B //2^( 22 /128-1) - 2^(- 22 /128-1), 2^(- 22 /128-1) + .quad 0x3FBFF79F0BEFA2C7, 0x3FDC40AB5FFFD07A //2^( 23 /128-1) - 2^(- 23 /128-1), 2^(- 23 /128-1) + .quad 0x3FC0AEB1FECAE3A9, 0x3FDC199BDD85529C //2^( 24 /128-1) - 2^(- 24 /128-1), 2^(- 24 /128-1) + .quad 0x3FC161B4871C5CEC, 0x3FDBF2C25BD71E09 //2^( 25 /128-1) - 2^(- 25 /128-1), 2^(- 25 /128-1) + .quad 0x3FC214D876F26FD0, 0x3FDBCC1E904BC1D2 //2^( 26 /128-1) - 2^(- 26 /128-1), 2^(- 26 /128-1) + .quad 0x3FC2C81F2693816F, 0x3FDBA5B030A1064A //2^( 27 /128-1) - 2^(- 27 /128-1), 2^(- 27 /128-1) + .quad 0x3FC37B89EE88BEF7, 0x3FDB7F76F2FB5E47 //2^( 28 /128-1) - 2^(- 28 /128-1), 2^(- 28 /128-1) + .quad 0x3FC42F1A27A0B3CD, 0x3FDB59728DE5593A //2^( 29 /128-1) - 2^(- 29 /128-1), 2^(- 29 /128-1) + .quad 0x3FC4E2D12AF1E037, 0x3FDB33A2B84F15FB //2^( 30 /128-1) - 2^(- 30 /128-1), 2^(- 30 /128-1) + .quad 0x3FC596B051DD508D, 0x3FDB0E07298DB666 //2^( 31 /128-1) - 2^(- 31 /128-1), 2^(- 31 /128-1) + .quad 0x3FC64AB8F61134FA, 0x3FDAE89F995AD3AD //2^( 32 /128-1) - 2^(- 32 /128-1), 2^(- 32 /128-1) + .quad 0x3FC6FEEC718B79D1, 0x3FDAC36BBFD3F37A //2^( 33 /128-1) - 2^(- 33 /128-1), 2^(- 33 /128-1) + .quad 0x3FC7B34C1E9C607F, 0x3FDA9E6B5579FDBF //2^( 34 /128-1) - 2^(- 34 /128-1), 2^(- 34 /128-1) + .quad 0x3FC867D957E91912, 0x3FDA799E1330B358 //2^( 35 /128-1) - 2^(- 35 /128-1), 2^(- 35 /128-1) + .quad 0x3FC91C95786E5C72, 0x3FDA5503B23E255D //2^( 36 /128-1) - 2^(- 36 /128-1), 2^(- 36 /128-1) + .quad 0x3FC9D181DB83072F, 0x3FDA309BEC4A2D33 //2^( 37 /128-1) - 2^(- 37 /128-1), 2^(- 37 /128-1) + .quad 0x3FCA869FDCDAB512, 0x3FDA0C667B5DE565 //2^( 38 /128-1) - 2^(- 38 /128-1), 2^(- 38 /128-1) + .quad 0x3FCB3BF0D8885D4C, 0x3FD9E86319E32323 //2^( 39 /128-1) - 2^(- 39 /128-1), 2^(- 39 /128-1) + .quad 0x3FCBF1762B00EF69, 0x3FD9C49182A3F090 //2^( 40 /128-1) - 2^(- 40 /128-1), 2^(- 40 /128-1) + .quad 0x3FCCA731311DF0FB, 0x3FD9A0F170CA07BA //2^( 41 /128-1) - 2^(- 41 /128-1), 2^(- 41 /128-1) + .quad 0x3FCD5D2348201C09, 0x3FD97D829FDE4E50 //2^( 42 /128-1) - 2^(- 42 /128-1), 2^(- 42 /128-1) + .quad 0x3FCE134DCDB1FE3E, 0x3FD95A44CBC8520F //2^( 43 /128-1) - 2^(- 43 /128-1), 2^(- 43 /128-1) + .quad 0x3FCEC9B21FEA98EA, 0x3FD93737B0CDC5E5 //2^( 44 /128-1) - 2^(- 44 /128-1), 2^(- 44 /128-1) + .quad 0x3FCF80519D5001D3, 0x3FD9145B0B91FFC6 //2^( 45 /128-1) - 2^(- 45 /128-1), 2^(- 45 /128-1) + .quad 0x3FD01B96D26D026A, 0x3FD8F1AE99157736 //2^( 46 /128-1) - 2^(- 46 /128-1), 2^(- 46 /128-1) + .quad 0x3FD07723CAFA6331, 0x3FD8CF3216B5448C //2^( 47 /128-1) - 2^(- 47 /128-1), 2^(- 47 /128-1) + .quad 0x3FD0D2D06841B373, 0x3FD8ACE5422AA0DB //2^( 48 /128-1) - 2^(- 48 /128-1), 2^(- 48 /128-1) + .quad 0x3FD12E9D5A715381, 0x3FD88AC7D98A6699 //2^( 49 /128-1) - 2^(- 49 /128-1), 2^(- 49 /128-1) + .quad 0x3FD18A8B51F5C661, 0x3FD868D99B4492ED //2^( 50 /128-1) - 2^(- 50 /128-1), 2^(- 50 /128-1) + .quad 0x3FD1E69AFF7B04D7, 0x3FD8471A4623C7AD //2^( 51 /128-1) - 2^(- 51 /128-1), 2^(- 51 /128-1) + .quad 0x3FD242CD13EDD0F1, 0x3FD82589994CCE13 //2^( 52 /128-1) - 2^(- 52 /128-1), 2^(- 52 /128-1) + .quad 0x3FD29F22407D0A0C, 0x3FD80427543E1A12 //2^( 53 /128-1) - 2^(- 53 /128-1), 2^(- 53 /128-1) + .quad 0x3FD2FB9B369B0153, 0x3FD7E2F336CF4E62 //2^( 54 /128-1) - 2^(- 54 /128-1), 2^(- 54 /128-1) + .quad 0x3FD35838A7FECEC8, 0x3FD7C1ED0130C132 //2^( 55 /128-1) - 2^(- 55 /128-1), 2^(- 55 /128-1) + .quad 0x3FD3B4FB46A5A6CC, 0x3FD7A11473EB0187 //2^( 56 /128-1) - 2^(- 56 /128-1), 2^(- 56 /128-1) + .quad 0x3FD411E3C4D4302F, 0x3FD780694FDE5D3F //2^( 57 /128-1) - 2^(- 57 /128-1), 2^(- 57 /128-1) + .quad 0x3FD46EF2D517DAC8, 0x3FD75FEB564267C9 //2^( 58 /128-1) - 2^(- 58 /128-1), 2^(- 58 /128-1) + .quad 0x3FD4CC292A48369E, 0x3FD73F9A48A58174 //2^( 59 /128-1) - 2^(- 59 /128-1), 2^(- 59 /128-1) + .quad 0x3FD5298777884B96, 0x3FD71F75E8EC5F74 //2^( 60 /128-1) - 2^(- 60 /128-1), 2^(- 60 /128-1) + .quad 0x3FD5870E7047F1BC, 0x3FD6FF7DF9519484 //2^( 61 /128-1) - 2^(- 61 /128-1), 2^(- 61 /128-1) + .quad 0x3FD5E4BEC8452A1A, 0x3FD6DFB23C651A2F //2^( 62 /128-1) - 2^(- 62 /128-1), 2^(- 62 /128-1) + .quad 0x3FD64299338D7827, 0x3FD6C012750BDABF //2^( 63 /128-1) - 2^(- 63 /128-1), 2^(- 63 /128-1) + .quad 0x3FD6A09E667F3BCD, 0x3FD6A09E667F3BCD //2^( 64 /128-1) - 2^(- 64 /128-1), 2^(- 64 /128-1) + .quad 0x3FD6FECF15CB0C0B, 0x3FD68155D44CA973 //2^( 65 /128-1) - 2^(- 65 /128-1), 2^(- 65 /128-1) + .quad 0x3FD75D2BF6751239, 0x3FD6623882552225 //2^( 66 /128-1) - 2^(- 66 /128-1), 2^(- 66 /128-1) + .quad 0x3FD7BBB5BDD665E8, 0x3FD6434634CCC320 //2^( 67 /128-1) - 2^(- 67 /128-1), 2^(- 67 /128-1) + .quad 0x3FD81A6D219E6963, 0x3FD6247EB03A5585 //2^( 68 /128-1) - 2^(- 68 /128-1), 2^(- 68 /128-1) + .quad 0x3FD87952D7D426DF, 0x3FD605E1B976DC09 //2^( 69 /128-1) - 2^(- 69 /128-1), 2^(- 69 /128-1) + .quad 0x3FD8D86796D7AE49, 0x3FD5E76F15AD2148 //2^( 70 /128-1) - 2^(- 70 /128-1), 2^(- 70 /128-1) + .quad 0x3FD937AC156373C8, 0x3FD5C9268A5946B7 //2^( 71 /128-1) - 2^(- 71 /128-1), 2^(- 71 /128-1) + .quad 0x3FD997210A8DAEE4, 0x3FD5AB07DD485429 //2^( 72 /128-1) - 2^(- 72 /128-1), 2^(- 72 /128-1) + .quad 0x3FD9F6C72DC9BA68, 0x3FD58D12D497C7FD //2^( 73 /128-1) - 2^(- 73 /128-1), 2^(- 73 /128-1) + .quad 0x3FDA569F36E974EA, 0x3FD56F4736B527DA //2^( 74 /128-1) - 2^(- 74 /128-1), 2^(- 74 /128-1) + .quad 0x3FDAB6A9DE1EA215, 0x3FD551A4CA5D920F //2^( 75 /128-1) - 2^(- 75 /128-1), 2^(- 75 /128-1) + .quad 0x3FDB16E7DBFC4CA3, 0x3FD5342B569D4F82 //2^( 76 /128-1) - 2^(- 76 /128-1), 2^(- 76 /128-1) + .quad 0x3FDB7759E9782918, 0x3FD516DAA2CF6642 //2^( 77 /128-1) - 2^(- 77 /128-1), 2^(- 77 /128-1) + .quad 0x3FDBD800BFEBF932, 0x3FD4F9B2769D2CA7 //2^( 78 /128-1) - 2^(- 78 /128-1), 2^(- 78 /128-1) + .quad 0x3FDC38DD1916F025, 0x3FD4DCB299FDDD0D //2^( 79 /128-1) - 2^(- 79 /128-1), 2^(- 79 /128-1) + .quad 0x3FDC99EFAF1F1790, 0x3FD4BFDAD5362A27 //2^( 80 /128-1) - 2^(- 80 /128-1), 2^(- 80 /128-1) + .quad 0x3FDCFB393C92B539, 0x3FD4A32AF0D7D3DE //2^( 81 /128-1) - 2^(- 81 /128-1), 2^(- 81 /128-1) + .quad 0x3FDD5CBA7C69B19C, 0x3FD486A2B5C13CD0 //2^( 82 /128-1) - 2^(- 82 /128-1), 2^(- 82 /128-1) + .quad 0x3FDDBE742A06FF34, 0x3FD46A41ED1D0057 //2^( 83 /128-1) - 2^(- 83 /128-1), 2^(- 83 /128-1) + .quad 0x3FDE2067013A029D, 0x3FD44E086061892D //2^( 84 /128-1) - 2^(- 84 /128-1), 2^(- 84 /128-1) + .quad 0x3FDE8293BE3FFB87, 0x3FD431F5D950A897 //2^( 85 /128-1) - 2^(- 85 /128-1), 2^(- 85 /128-1) + .quad 0x3FDEE4FB1DC56E75, 0x3FD4160A21F72E2A //2^( 86 /128-1) - 2^(- 86 /128-1), 2^(- 86 /128-1) + .quad 0x3FDF479DDCE78F58, 0x3FD3FA4504AC801C //2^( 87 /128-1) - 2^(- 87 /128-1), 2^(- 87 /128-1) + .quad 0x3FDFAA7CB935ACFE, 0x3FD3DEA64C123422 //2^( 88 /128-1) - 2^(- 88 /128-1), 2^(- 88 /128-1) + .quad 0x3FE006CC38594EB1, 0x3FD3C32DC313A8E5 //2^( 89 /128-1) - 2^(- 89 /128-1), 2^(- 89 /128-1) + .quad 0x3FE03878E0EB1569, 0x3FD3A7DB34E59FF7 //2^( 90 /128-1) - 2^(- 90 /128-1), 2^(- 90 /128-1) + .quad 0x3FE06A44B5C74101, 0x3FD38CAE6D05D866 //2^( 91 /128-1) - 2^(- 91 /128-1), 2^(- 91 /128-1) + .quad 0x3FE09C3016A0D077, 0x3FD371A7373AA9CB //2^( 92 /128-1) - 2^(- 92 /128-1), 2^(- 92 /128-1) + .quad 0x3FE0CE3B63676360, 0x3FD356C55F929FF1 //2^( 93 /128-1) - 2^(- 93 /128-1), 2^(- 93 /128-1) + .quad 0x3FE10066FC47F240, 0x3FD33C08B26416FF //2^( 94 /128-1) - 2^(- 94 /128-1), 2^(- 94 /128-1) + .quad 0x3FE132B341AD8761, 0x3FD32170FC4CD831 //2^( 95 /128-1) - 2^(- 95 /128-1), 2^(- 95 /128-1) + .quad 0x3FE165209441F823, 0x3FD306FE0A31B715 //2^( 96 /128-1) - 2^(- 96 /128-1), 2^(- 96 /128-1) + .quad 0x3FE197AF54EE9EBB, 0x3FD2ECAFA93E2F56 //2^( 97 /128-1) - 2^(- 97 /128-1), 2^(- 97 /128-1) + .quad 0x3FE1CA5FE4DD1475, 0x3FD2D285A6E4030B //2^( 98 /128-1) - 2^(- 98 /128-1), 2^(- 98 /128-1) + .quad 0x3FE1FD32A577EC72, 0x3FD2B87FD0DAD990 //2^( 99 /128-1) - 2^(- 99 /128-1), 2^(- 99 /128-1) + .quad 0x3FE23027F86B6ED6, 0x3FD29E9DF51FDEE1 //2^( 100 /128-1) - 2^(- 100 /128-1), 2^(- 100 /128-1) + .quad 0x3FE263403FA65489, 0x3FD284DFE1F56381 //2^( 101 /128-1) - 2^(- 101 /128-1), 2^(- 101 /128-1) + .quad 0x3FE2967BDD5A8364, 0x3FD26B4565E27CDD //2^( 102 /128-1) - 2^(- 102 /128-1), 2^(- 102 /128-1) + .quad 0x3FE2C9DB33FDCAE9, 0x3FD251CE4FB2A63F //2^( 103 /128-1) - 2^(- 103 /128-1), 2^(- 103 /128-1) + .quad 0x3FE2FD5EA64AA180, 0x3FD2387A6E756238 //2^( 104 /128-1) - 2^(- 104 /128-1), 2^(- 104 /128-1) + .quad 0x3FE331069740E22F, 0x3FD21F49917DDC96 //2^( 105 /128-1) - 2^(- 105 /128-1), 2^(- 105 /128-1) + .quad 0x3FE364D36A268AE0, 0x3FD2063B88628CD6 //2^( 106 /128-1) - 2^(- 106 /128-1), 2^(- 106 /128-1) + .quad 0x3FE398C582887B27, 0x3FD1ED5022FCD91D //2^( 107 /128-1) - 2^(- 107 /128-1), 2^(- 107 /128-1) + .quad 0x3FE3CCDD443B3394, 0x3FD1D4873168B9AA //2^( 108 /128-1) - 2^(- 108 /128-1), 2^(- 108 /128-1) + .quad 0x3FE4011B135B9590, 0x3FD1BBE084045CD4 //2^( 109 /128-1) - 2^(- 109 /128-1), 2^(- 109 /128-1) + .quad 0x3FE4357F544FA3C1, 0x3FD1A35BEB6FCB75 //2^( 110 /128-1) - 2^(- 110 /128-1), 2^(- 110 /128-1) + .quad 0x3FE46A0A6BC742FD, 0x3FD18AF9388C8DEA //2^( 111 /128-1) - 2^(- 111 /128-1), 2^(- 111 /128-1) + .quad 0x3FE49EBCBEBCFBCA, 0x3FD172B83C7D517B //2^( 112 /128-1) - 2^(- 112 /128-1), 2^(- 112 /128-1) + .quad 0x3FE4D396B276BC6F, 0x3FD15A98C8A58E51 //2^( 113 /128-1) - 2^(- 113 /128-1), 2^(- 113 /128-1) + .quad 0x3FE50898AC869B96, 0x3FD1429AAEA92DE0 //2^( 114 /128-1) - 2^(- 114 /128-1), 2^(- 114 /128-1) + .quad 0x3FE53DC312CB9B7A, 0x3FD12ABDC06C31CC //2^( 115 /128-1) - 2^(- 115 /128-1), 2^(- 115 /128-1) + .quad 0x3FE573164B726DB6, 0x3FD11301D0125B51 //2^( 116 /128-1) - 2^(- 116 /128-1), 2^(- 116 /128-1) + .quad 0x3FE5A892BCF6379B, 0x3FD0FB66AFFED31B //2^( 117 /128-1) - 2^(- 117 /128-1), 2^(- 117 /128-1) + .quad 0x3FE5DE38CE215725, 0x3FD0E3EC32D3D1A2 //2^( 118 /128-1) - 2^(- 118 /128-1), 2^(- 118 /128-1) + .quad 0x3FE61408E60E2888, 0x3FD0CC922B7247F7 //2^( 119 /128-1) - 2^(- 119 /128-1), 2^(- 119 /128-1) + .quad 0x3FE64A036C27CC52, 0x3FD0B5586CF9890F //2^( 120 /128-1) - 2^(- 120 /128-1), 2^(- 120 /128-1) + .quad 0x3FE68028C82AEE2F, 0x3FD09E3ECAC6F383 //2^( 121 /128-1) - 2^(- 121 /128-1), 2^(- 121 /128-1) + .quad 0x3FE6B67962268C43, 0x3FD0874518759BC8 //2^( 122 /128-1) - 2^(- 122 /128-1), 2^(- 122 /128-1) + .quad 0x3FE6ECF5A27CBF28, 0x3FD0706B29DDF6DE //2^( 123 /128-1) - 2^(- 123 /128-1), 2^(- 123 /128-1) + .quad 0x3FE7239DF1E38286, 0x3FD059B0D3158574 //2^( 124 /128-1) - 2^(- 124 /128-1), 2^(- 124 /128-1) + .quad 0x3FE75A72B9657E51, 0x3FD04315E86E7F85 //2^( 125 /128-1) - 2^(- 125 /128-1), 2^(- 125 /128-1) + .quad 0x3FE791746262D0A8, 0x3FD02C9A3E778061 //2^( 126 /128-1) - 2^(- 126 /128-1), 2^(- 126 /128-1) + .quad 0x3FE7C8A35691D856, 0x3FD0163DA9FB3335 //2^( 127 /128-1) - 2^(- 127 /128-1), 2^(- 127 /128-1) + .align 64 + .quad 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000, 0x42C8000000000000 /* _dbShifter = 1.5 * 2^(52-k)*/ + .align 64 + .long 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99 /* _iDomainRange 0x40861d9ac12a3e85 =(1021*2^K-0.5)*log(2)/2^K -needed for quick exp*/ + .align 64 + .quad 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD /* _dPC2 */ + .align 64 + .quad 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD, 0x3FC55555555554AD /* _dPC3 */ + .align 64 + .quad 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299 /* _dPC4 */ + .align 64 + .quad 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425, 0x3F8111115712F425 /* _dPC5 */ + .align 64 + .quad 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f, 0x000000000000007f /* _lIndexMask */ + .align 64 + .type __svml_dsinh_data_internal,@object + .size __svml_dsinh_data_internal,.-__svml_dsinh_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core-avx2.S new file mode 100644 index 0000000000..06525b7b37 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core-avx2.S @@ -0,0 +1,20 @@ +/* AVX2 version of vectorized sinhf. + 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 + . */ + +#define _ZGVeN16v_sinhf _ZGVeN16v_sinhf_avx2_wrapper +#include "../svml_s_sinhf16_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core.c new file mode 100644 index 0000000000..6a954caa37 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized sinhf, 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 + . */ + +#define SYMBOL_NAME _ZGVeN16v_sinhf +#include "ifunc-mathvec-avx512-skx.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVeN16v_sinhf, __GI__ZGVeN16v_sinhf, + __redirect__ZGVeN16v_sinhf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core_avx512.S new file mode 100644 index 0000000000..1119c00259 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf16_core_avx512.S @@ -0,0 +1,318 @@ +/* Function sinhf 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_ssinh_data_internal + */ +#define _sInvLn2 0 +#define _sLn2hi 64 +#define _sLn2lo 128 +#define _sSign 192 +#define _sShifter 256 +#define _iDomainRange 320 +#define _sPC1 384 +#define _sPC2 448 +#define _sPC3 512 +#define _sPC4 576 +#define _sPC5 640 +#define _sPC6 704 +#define _iHalf 768 + +#include + + .text + .section .text.exex512,"ax",@progbits +ENTRY(_ZGVeN16v_sinhf_skx) + pushq %rbp + cfi_def_cfa_offset(16) + movq %rsp, %rbp + cfi_def_cfa(6, 16) + cfi_offset(6, -16) + andq $-64, %rsp + subq $192, %rsp + vmovaps %zmm0, %zmm5 + +/* + * Implementation + * Abs argument + */ + vandps _sSign+__svml_ssinh_data_internal(%rip), %zmm5, %zmm4 + +/* + * Check for overflow\underflow + * MORE faster than GE? + */ + vpternlogd $255, %zmm6, %zmm6, %zmm6 + vmovups _sShifter+__svml_ssinh_data_internal(%rip), %zmm7 + +/* + * Load argument + * dM = x/log(2) + RShifter + */ + vmovups _sInvLn2+__svml_ssinh_data_internal(%rip), %zmm11 + vmovups _sLn2hi+__svml_ssinh_data_internal(%rip), %zmm8 + vmovups _sLn2lo+__svml_ssinh_data_internal(%rip), %zmm10 + vmovups _iHalf+__svml_ssinh_data_internal(%rip), %zmm12 + vmovups _sPC5+__svml_ssinh_data_internal(%rip), %zmm0 + vmovups _sPC6+__svml_ssinh_data_internal(%rip), %zmm3 + +/* x^2 */ + vmovups _sPC2+__svml_ssinh_data_internal(%rip), %zmm2 + vxorps %zmm5, %zmm4, %zmm1 + vfmadd213ps {rn-sae}, %zmm7, %zmm1, %zmm11 + vpcmpd $2, _iDomainRange+__svml_ssinh_data_internal(%rip), %zmm1, %k1 + +/* + * G1,G2 2^N,2^(-N) + * iM now is an EXP(2^N) + */ + vpslld $23, %zmm11, %zmm13 + +/* + * R + * sN = sM - RShifter + */ + vsubps {rn-sae}, %zmm7, %zmm11, %zmm9 + vpaddd %zmm13, %zmm12, %zmm14 + vpsubd %zmm13, %zmm12, %zmm15 + +/* sG1 = 2^(N-1)+2^(-N-1) */ + vaddps {rn-sae}, %zmm15, %zmm14, %zmm7 + vpandnd %zmm1, %zmm1, %zmm6{%k1} + +/* sR = sX - sN*Log2_hi */ + vfnmadd231ps {rn-sae}, %zmm8, %zmm9, %zmm1 + vptestmd %zmm6, %zmm6, %k0 + +/* sG2 = 2^(N-1)-2^(-N-1) */ + vsubps {rn-sae}, %zmm15, %zmm14, %zmm8 + +/* sR = (sX - sN*Log2_hi) - sN*Log2_lo */ + vfnmadd231ps {rn-sae}, %zmm10, %zmm9, %zmm1 + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*(a5+{v1 r^2*a7})))) = r + r*(r^2*(a3+r^2*(a5+r^2*a7))) .... + * sSinh_r = (a3+r^2*a5) + */ + vmovups _sPC3+__svml_ssinh_data_internal(%rip), %zmm14 + kmovw %k0, %edx + +/* sR2 = sR^2 */ + vmulps {rn-sae}, %zmm1, %zmm1, %zmm6 + vfmadd231ps {rn-sae}, %zmm6, %zmm0, %zmm14 + +/* sSinh_r = r^2*(a3+r^2*a5) */ + vmulps {rn-sae}, %zmm6, %zmm14, %zmm0 + +/* sSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + vfmadd213ps {rn-sae}, %zmm1, %zmm1, %zmm0 + +/* + * sinh(X) = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) + * sOut = (a4 +a6*sR2) + */ + vmovups _sPC4+__svml_ssinh_data_internal(%rip), %zmm1 + vfmadd231ps {rn-sae}, %zmm6, %zmm3, %zmm1 + +/* sOut = a2+sR2*(a4+a6*sR2) */ + vfmadd213ps {rn-sae}, %zmm2, %zmm6, %zmm1 + +/* sOut = sR2*(a2+sR2*(a4+a6*sR2) */ + vmulps {rn-sae}, %zmm6, %zmm1, %zmm2 + +/* sOut = sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vmulps {rn-sae}, %zmm8, %zmm2, %zmm3 + +/* sOut = sG1*sinh(dR)+sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vfmadd213ps {rn-sae}, %zmm3, %zmm0, %zmm7 + +/* sOut = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vaddps {rn-sae}, %zmm8, %zmm7, %zmm9 + +/* Ret H */ + vorps %zmm9, %zmm4, %zmm0 + testl %edx, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm5 + +/* 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): + vmovups %zmm5, 64(%rsp) + vmovups %zmm0, 128(%rsp) + # LOE rbx r12 r13 r14 r15 edx zmm0 + + 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: -176; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 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 128(%rsp), %zmm0 + +/* 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: -176; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 + # LOE rbx r12 r13 r14 r15 zmm0 + +/* Scalar math fucntion call + * to process special input + */ + +L(SCALAR_MATH_CALL): + movl %r12d, %r14d + movss 64(%rsp,%r14,4), %xmm0 + call sinhf@PLT + # LOE rbx r14 r15 r12d r13d xmm0 + + movss %xmm0, 128(%rsp,%r14,4) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx r15 r12d r13d +END(_ZGVeN16v_sinhf_skx) + + .section .rodata, "a" + .align 64 + +#ifdef __svml_ssinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(64)) VUINT32 _sInvLn2[16][1]; + __declspec(align(64)) VUINT32 _sLn2hi[16][1]; + __declspec(align(64)) VUINT32 _sLn2lo[16][1]; + __declspec(align(64)) VUINT32 _sSign[16][1]; + __declspec(align(64)) VUINT32 _sShifter[16][1]; + __declspec(align(64)) VUINT32 _iDomainRange[16][1]; + __declspec(align(64)) VUINT32 _sPC1[16][1]; + __declspec(align(64)) VUINT32 _sPC2[16][1]; + __declspec(align(64)) VUINT32 _sPC3[16][1]; + __declspec(align(64)) VUINT32 _sPC4[16][1]; + __declspec(align(64)) VUINT32 _sPC5[16][1]; + __declspec(align(64)) VUINT32 _sPC6[16][1]; + __declspec(align(64)) VUINT32 _iHalf[16][1]; +} __svml_ssinh_data_internal; +#endif +__svml_ssinh_data_internal: + .long 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B /* _sInvLn2 */ //k=0 + .align 64 + .long 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000 /* _sLn2hi */ + .align 64 + .long 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4 /* _sLn2lo */ + .align 64 + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 /* _sSign */ + .align 64 + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter */ + .align 64 + .long 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E /* _iDomainRange */ + .align 64 + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 /* _sPC1=1 */ + .align 64 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sPC2 */ + .align 64 + .long 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57 /* _sPC3 */ + .align 64 + .long 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72 /* _sPC4 */ + .align 64 + .long 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461 /* _sPC5 */ + .align 64 + .long 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3 /* _sPC6 */ + // Integer constants + .align 64 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _iHalf*/ + .align 64 + .type __svml_ssinh_data_internal,@object + .size __svml_ssinh_data_internal,.-__svml_ssinh_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core-sse2.S new file mode 100644 index 0000000000..1b31095fe1 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core-sse2.S @@ -0,0 +1,20 @@ +/* SSE2 version of vectorized sinhf, 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 + . */ + +#define _ZGVbN4v_sinhf _ZGVbN4v_sinhf_sse2 +#include "../svml_s_sinhf4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core.c new file mode 100644 index 0000000000..9d4297c2c9 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized sinhf, 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 + . */ + +#define SYMBOL_NAME _ZGVbN4v_sinhf +#include "ifunc-mathvec-sse4_1.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVbN4v_sinhf, __GI__ZGVbN4v_sinhf, + __redirect__ZGVbN4v_sinhf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core_sse4.S new file mode 100644 index 0000000000..82d6f55d33 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf4_core_sse4.S @@ -0,0 +1,308 @@ +/* Function sinhf 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_ssinh_data_internal + */ +#define _sInvLn2 0 +#define _sLn2hi 16 +#define _sLn2lo 32 +#define _sSign 48 +#define _sShifter 64 +#define _iDomainRange 80 +#define _sPC1 96 +#define _sPC2 112 +#define _sPC3 128 +#define _sPC4 144 +#define _sPC5 160 +#define _sPC6 176 +#define _iHalf 192 + +#include + + .text + .section .text.sse4,"ax",@progbits +ENTRY(_ZGVbN4v_sinhf_sse4) + subq $72, %rsp + cfi_def_cfa_offset(80) + +/* + * Implementation + * Abs argument + */ + movups _sSign+__svml_ssinh_data_internal(%rip), %xmm14 + andps %xmm0, %xmm14 + movaps %xmm14, %xmm10 + +/* + * Load argument + * dM = x/log(2) + RShifter + */ + movups _sInvLn2+__svml_ssinh_data_internal(%rip), %xmm7 + pxor %xmm0, %xmm10 + mulps %xmm10, %xmm7 + +/* + * Check for overflow\underflow + * MORE faster than GE? + */ + movaps %xmm10, %xmm1 + movups _sShifter+__svml_ssinh_data_internal(%rip), %xmm2 + +/* sR = sX - sN*Log2_hi */ + movups _sLn2hi+__svml_ssinh_data_internal(%rip), %xmm3 + addps %xmm2, %xmm7 + +/* + * R + * sN = sM - RShifter + */ + movaps %xmm7, %xmm4 + +/* + * G1,G2 2^N,2^(-N) + * iM now is an EXP(2^N) + */ + pslld $23, %xmm7 + +/* sR = (sX - sN*Log2_hi) - sN*Log2_lo */ + movups _sLn2lo+__svml_ssinh_data_internal(%rip), %xmm5 + subps %xmm2, %xmm4 + mulps %xmm4, %xmm3 + mulps %xmm4, %xmm5 + subps %xmm3, %xmm10 + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*(a5+{v1 r^2*a7})))) = r + r*(r^2*(a3+r^2*(a5+r^2*a7))) .... + * sSinh_r = (a3+r^2*a5) + */ + movups _sPC5+__svml_ssinh_data_internal(%rip), %xmm8 + subps %xmm5, %xmm10 + +/* sR2 = sR^2 */ + movaps %xmm10, %xmm12 + mulps %xmm10, %xmm12 + +/* + * sinh(X) = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) + * sOut = (a4 +a6*sR2) + */ + movups _sPC6+__svml_ssinh_data_internal(%rip), %xmm9 + mulps %xmm12, %xmm8 + mulps %xmm12, %xmm9 + addps _sPC3+__svml_ssinh_data_internal(%rip), %xmm8 + addps _sPC4+__svml_ssinh_data_internal(%rip), %xmm9 + +/* sSinh_r = r^2*(a3+r^2*a5) */ + mulps %xmm12, %xmm8 + +/* sOut = a2+sR2*(a4+a6*sR2) */ + mulps %xmm12, %xmm9 + +/* sSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + mulps %xmm10, %xmm8 + addps _sPC2+__svml_ssinh_data_internal(%rip), %xmm9 + addps %xmm8, %xmm10 + +/* sOut = sR2*(a2+sR2*(a4+a6*sR2) */ + mulps %xmm9, %xmm12 + movdqu _iHalf+__svml_ssinh_data_internal(%rip), %xmm6 + movdqa %xmm6, %xmm13 + psubd %xmm7, %xmm6 + paddd %xmm7, %xmm13 + +/* sG1 = 2^(N-1)+2^(-N-1) */ + movdqa %xmm13, %xmm11 + +/* sG2 = 2^(N-1)-2^(-N-1) */ + subps %xmm6, %xmm13 + addps %xmm6, %xmm11 + +/* sOut = sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + mulps %xmm13, %xmm12 + +/* sOut = sG1*sinh(dR)+sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + mulps %xmm10, %xmm11 + pcmpgtd _iDomainRange+__svml_ssinh_data_internal(%rip), %xmm1 + addps %xmm11, %xmm12 + movmskps %xmm1, %edx + +/* sOut = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + addps %xmm12, %xmm13 + +/* Ret H */ + orps %xmm13, %xmm14 + testl %edx, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm14 + +/* Restore registers + * and exit the function + */ + +L(EXIT): + movaps %xmm14, %xmm0 + addq $72, %rsp + cfi_def_cfa_offset(8) + ret + cfi_def_cfa_offset(80) + +/* Branch to process + * special inputs + */ + +L(SPECIAL_VALUES_BRANCH): + movups %xmm0, 32(%rsp) + movups %xmm14, 48(%rsp) + # LOE rbx rbp r12 r13 r14 r15 edx + + xorl %eax, %eax + movq %r12, 16(%rsp) + cfi_offset(12, -64) + movl %eax, %r12d + movq %r13, 8(%rsp) + cfi_offset(13, -72) + movl %edx, %r13d + movq %r14, (%rsp) + cfi_offset(14, -80) + # 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 48(%rsp), %xmm14 + +/* Go to exit */ + jmp L(EXIT) + cfi_offset(12, -64) + cfi_offset(13, -72) + cfi_offset(14, -80) + # LOE rbx rbp r12 r13 r14 r15 xmm14 + +/* Scalar math fucntion call + * to process special input + */ + +L(SCALAR_MATH_CALL): + movl %r12d, %r14d + movss 32(%rsp,%r14,4), %xmm0 + call sinhf@PLT + # LOE rbx rbp r14 r15 r12d r13d xmm0 + + movss %xmm0, 48(%rsp,%r14,4) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx rbp r15 r12d r13d +END(_ZGVbN4v_sinhf_sse4) + + .section .rodata, "a" + .align 16 + +#ifdef __svml_ssinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(16)) VUINT32 _sInvLn2[4][1]; + __declspec(align(16)) VUINT32 _sLn2hi[4][1]; + __declspec(align(16)) VUINT32 _sLn2lo[4][1]; + __declspec(align(16)) VUINT32 _sSign[4][1]; + __declspec(align(16)) VUINT32 _sShifter[4][1]; + __declspec(align(16)) VUINT32 _iDomainRange[4][1]; + __declspec(align(16)) VUINT32 _sPC1[4][1]; + __declspec(align(16)) VUINT32 _sPC2[4][1]; + __declspec(align(16)) VUINT32 _sPC3[4][1]; + __declspec(align(16)) VUINT32 _sPC4[4][1]; + __declspec(align(16)) VUINT32 _sPC5[4][1]; + __declspec(align(16)) VUINT32 _sPC6[4][1]; + __declspec(align(16)) VUINT32 _iHalf[4][1]; +} __svml_ssinh_data_internal; +#endif +__svml_ssinh_data_internal: + .long 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B /* _sInvLn2 */ //k=0 + .align 16 + .long 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000 /* _sLn2hi */ + .align 16 + .long 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4 /* _sLn2lo */ + .align 16 + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000 /* _sSign */ + .align 16 + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter */ + .align 16 + .long 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E /* _iDomainRange */ + .align 16 + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 /* _sPC1=1 */ + .align 16 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sPC2 */ + .align 16 + .long 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57 /* _sPC3 */ + .align 16 + .long 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72 /* _sPC4 */ + .align 16 + .long 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461 /* _sPC5 */ + .align 16 + .long 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3 /* _sPC6 */ + // Integer constants + .align 16 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _iHalf*/ + .align 16 + .type __svml_ssinh_data_internal,@object + .size __svml_ssinh_data_internal,.-__svml_ssinh_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core-sse.S new file mode 100644 index 0000000000..d3c9c607a0 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core-sse.S @@ -0,0 +1,20 @@ +/* SSE version of vectorized sinhf, 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 + . */ + +#define _ZGVdN8v_sinhf _ZGVdN8v_sinhf_sse_wrapper +#include "../svml_s_sinhf8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core.c new file mode 100644 index 0000000000..2a2e21e742 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized sinhf, 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 + . */ + +#define SYMBOL_NAME _ZGVdN8v_sinhf +#include "ifunc-mathvec-avx2.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVdN8v_sinhf, __GI__ZGVdN8v_sinhf, + __redirect__ZGVdN8v_sinhf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core_avx2.S new file mode 100644 index 0000000000..ea13fb60d4 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinhf8_core_avx2.S @@ -0,0 +1,309 @@ +/* Function sinhf 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: + * + * Compute sinh(x) as (exp(x)-exp(-x))/2, + * where exp is calculated as + * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r) + * + * Special cases: + * + * sinh(NaN) = quiet NaN, and raise invalid exception + * sinh(INF) = that INF + * sinh(x) = x for subnormals + * sinh(x) overflows for big x and returns MAXLOG+log(2) + * + */ + +/* Offsets for data table __svml_ssinh_data_internal + */ +#define _sInvLn2 0 +#define _sLn2hi 32 +#define _sLn2lo 64 +#define _sSign 96 +#define _sShifter 128 +#define _iDomainRange 160 +#define _sPC1 192 +#define _sPC2 224 +#define _sPC3 256 +#define _sPC4 288 +#define _sPC5 320 +#define _sPC6 352 +#define _iHalf 384 + +#include + + .text + .section .text.avx2,"ax",@progbits +ENTRY(_ZGVdN8v_sinhf_avx2) + pushq %rbp + cfi_def_cfa_offset(16) + movq %rsp, %rbp + cfi_def_cfa(6, 16) + cfi_offset(6, -16) + andq $-32, %rsp + subq $96, %rsp + vmovups _sInvLn2+__svml_ssinh_data_internal(%rip), %ymm7 + vmovups _sShifter+__svml_ssinh_data_internal(%rip), %ymm4 + vmovups _sLn2hi+__svml_ssinh_data_internal(%rip), %ymm5 + +/* + * sinh(X) = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) + * sOut = (a4 +a6*sR2) + */ + vmovups _sPC6+__svml_ssinh_data_internal(%rip), %ymm14 + +/* + * sinh(r) = r*((a1=1)+r^2*(a3+r^2*(a5+{v1 r^2*a7})))) = r + r*(r^2*(a3+r^2*(a5+r^2*a7))) .... + * sSinh_r = (a3+r^2*a5) + */ + vmovups _sPC5+__svml_ssinh_data_internal(%rip), %ymm12 + vmovups _iHalf+__svml_ssinh_data_internal(%rip), %ymm8 + vmovaps %ymm0, %ymm2 + +/* + * Implementation + * Abs argument + */ + vandps _sSign+__svml_ssinh_data_internal(%rip), %ymm2, %ymm1 + vxorps %ymm2, %ymm1, %ymm0 + +/* + * Load argument + * dM = x/log(2) + RShifter + */ + vfmadd213ps %ymm4, %ymm0, %ymm7 + +/* + * R + * sN = sM - RShifter + */ + vsubps %ymm4, %ymm7, %ymm6 + +/* + * G1,G2 2^N,2^(-N) + * iM now is an EXP(2^N) + */ + vpslld $23, %ymm7, %ymm9 + +/* + * Check for overflow\underflow + * MORE faster than GE? + */ + vpcmpgtd _iDomainRange+__svml_ssinh_data_internal(%rip), %ymm0, %ymm3 + +/* sR = sX - sN*Log2_hi */ + vfnmadd231ps %ymm5, %ymm6, %ymm0 + vpaddd %ymm9, %ymm8, %ymm10 + vpsubd %ymm9, %ymm8, %ymm11 + +/* sR = (sX - sN*Log2_hi) - sN*Log2_lo */ + vfnmadd231ps _sLn2lo+__svml_ssinh_data_internal(%rip), %ymm6, %ymm0 + +/* sR2 = sR^2 */ + vmulps %ymm0, %ymm0, %ymm13 + vfmadd213ps _sPC4+__svml_ssinh_data_internal(%rip), %ymm13, %ymm14 + vfmadd213ps _sPC3+__svml_ssinh_data_internal(%rip), %ymm13, %ymm12 + +/* sOut = a2+sR2*(a4+a6*sR2) */ + vfmadd213ps _sPC2+__svml_ssinh_data_internal(%rip), %ymm13, %ymm14 + +/* sSinh_r = r^2*(a3+r^2*a5) */ + vmulps %ymm12, %ymm13, %ymm12 + +/* sOut = sR2*(a2+sR2*(a4+a6*sR2) */ + vmulps %ymm14, %ymm13, %ymm15 + +/* sSinh_r = r + r*(r^2*(a3+r^2*a5)) */ + vfmadd213ps %ymm0, %ymm0, %ymm12 + vmovmskps %ymm3, %edx + +/* sG1 = 2^(N-1)+2^(-N-1) */ + vaddps %ymm11, %ymm10, %ymm3 + +/* sG2 = 2^(N-1)-2^(-N-1) */ + vsubps %ymm11, %ymm10, %ymm10 + +/* sOut = sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vmulps %ymm15, %ymm10, %ymm0 + +/* sOut = sG1*sinh(dR)+sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vfmadd213ps %ymm0, %ymm12, %ymm3 + +/* sOut = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) */ + vaddps %ymm3, %ymm10, %ymm4 + +/* Ret H */ + vorps %ymm4, %ymm1, %ymm0 + testl %edx, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm2 + +/* 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): + vmovups %ymm2, 32(%rsp) + vmovups %ymm0, 64(%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: -80; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 64(%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: -80; DW_OP_plus) */ + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 + movss 32(%rsp,%r14,4), %xmm0 + call sinhf@PLT + # LOE rbx r14 r15 r12d r13d xmm0 + + movss %xmm0, 64(%rsp,%r14,4) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx r15 r12d r13d +END(_ZGVdN8v_sinhf_avx2) + + .section .rodata, "a" + .align 32 + +#ifdef __svml_ssinh_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(32)) VUINT32 _sInvLn2[8][1]; + __declspec(align(32)) VUINT32 _sLn2hi[8][1]; + __declspec(align(32)) VUINT32 _sLn2lo[8][1]; + __declspec(align(32)) VUINT32 _sSign[8][1]; + __declspec(align(32)) VUINT32 _sShifter[8][1]; + __declspec(align(32)) VUINT32 _iDomainRange[8][1]; + __declspec(align(32)) VUINT32 _sPC1[8][1]; + __declspec(align(32)) VUINT32 _sPC2[8][1]; + __declspec(align(32)) VUINT32 _sPC3[8][1]; + __declspec(align(32)) VUINT32 _sPC4[8][1]; + __declspec(align(32)) VUINT32 _sPC5[8][1]; + __declspec(align(32)) VUINT32 _sPC6[8][1]; + __declspec(align(32)) VUINT32 _iHalf[8][1]; +} __svml_ssinh_data_internal; +#endif +__svml_ssinh_data_internal: + .long 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B /* _sInvLn2 */ //k=0 + .align 32 + .long 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000 /* _sLn2hi */ + .align 32 + .long 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4, 0x3805FDF4 /* _sLn2lo */ + .align 32 + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 /* _sSign */ + .align 32 + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter */ + .align 32 + .long 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E /* _iDomainRange */ + .align 32 + .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000 /* _sPC1=1 */ + .align 32 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sPC2 */ + .align 32 + .long 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57 /* _sPC3 */ + .align 32 + .long 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72 /* _sPC4 */ + .align 32 + .long 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461 /* _sPC5 */ + .align 32 + .long 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3 /* _sPC6 */ + // Integer constants + .align 32 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _iHalf*/ + .align 32 + .type __svml_ssinh_data_internal,@object + .size __svml_ssinh_data_internal,.-__svml_ssinh_data_internal diff --git a/sysdeps/x86_64/fpu/svml_d_sinh2_core.S b/sysdeps/x86_64/fpu/svml_d_sinh2_core.S new file mode 100644 index 0000000000..91bda7318c --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_sinh2_core.S @@ -0,0 +1,29 @@ +/* Function sinh 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 + . */ + +#include +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVbN2v_sinh) +WRAPPER_IMPL_SSE2 sinh +END (_ZGVbN2v_sinh) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVbN2v_sinh) +#endif diff --git a/sysdeps/x86_64/fpu/svml_d_sinh4_core.S b/sysdeps/x86_64/fpu/svml_d_sinh4_core.S new file mode 100644 index 0000000000..7b8091946a --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_sinh4_core.S @@ -0,0 +1,29 @@ +/* Function sinh 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 + . */ + +#include +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVdN4v_sinh) +WRAPPER_IMPL_AVX _ZGVbN2v_sinh +END (_ZGVdN4v_sinh) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVdN4v_sinh) +#endif diff --git a/sysdeps/x86_64/fpu/svml_d_sinh4_core_avx.S b/sysdeps/x86_64/fpu/svml_d_sinh4_core_avx.S new file mode 100644 index 0000000000..f773bf110c --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_sinh4_core_avx.S @@ -0,0 +1,25 @@ +/* Function sinh 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 + . */ + +#include +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVcN4v_sinh) +WRAPPER_IMPL_AVX _ZGVbN2v_sinh +END (_ZGVcN4v_sinh) diff --git a/sysdeps/x86_64/fpu/svml_d_sinh8_core.S b/sysdeps/x86_64/fpu/svml_d_sinh8_core.S new file mode 100644 index 0000000000..153a18429c --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_sinh8_core.S @@ -0,0 +1,25 @@ +/* Function sinh vectorized with AVX-512, wrapper to 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 + . */ + +#include +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVeN8v_sinh) +WRAPPER_IMPL_AVX512 _ZGVdN4v_sinh +END (_ZGVeN8v_sinh) diff --git a/sysdeps/x86_64/fpu/svml_s_sinhf16_core.S b/sysdeps/x86_64/fpu/svml_s_sinhf16_core.S new file mode 100644 index 0000000000..f8dc7da336 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_sinhf16_core.S @@ -0,0 +1,25 @@ +/* Function sinhf 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 + . */ + +#include +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVeN16v_sinhf) +WRAPPER_IMPL_AVX512 _ZGVdN8v_sinhf +END (_ZGVeN16v_sinhf) diff --git a/sysdeps/x86_64/fpu/svml_s_sinhf4_core.S b/sysdeps/x86_64/fpu/svml_s_sinhf4_core.S new file mode 100644 index 0000000000..d065d03eb6 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_sinhf4_core.S @@ -0,0 +1,29 @@ +/* Function sinhf vectorized with SSE2, 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 + . */ + +#include +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVbN4v_sinhf) +WRAPPER_IMPL_SSE2 sinhf +END (_ZGVbN4v_sinhf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVbN4v_sinhf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_sinhf8_core.S b/sysdeps/x86_64/fpu/svml_s_sinhf8_core.S new file mode 100644 index 0000000000..1194699a76 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_sinhf8_core.S @@ -0,0 +1,29 @@ +/* Function sinhf 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 + . */ + +#include +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVdN8v_sinhf) +WRAPPER_IMPL_AVX _ZGVbN4v_sinhf +END (_ZGVdN8v_sinhf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVdN8v_sinhf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_sinhf8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_sinhf8_core_avx.S new file mode 100644 index 0000000000..82c6b9b239 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_sinhf8_core_avx.S @@ -0,0 +1,25 @@ +/* Function sinhf 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 + . */ + +#include +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVcN8v_sinhf) +WRAPPER_IMPL_AVX _ZGVbN4v_sinhf +END (_ZGVcN8v_sinhf) diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx.c b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx.c new file mode 100644 index 0000000000..55aa36d866 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx.c @@ -0,0 +1 @@ +#include "test-double-libmvec-sinh.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx2.c b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx2.c new file mode 100644 index 0000000000..55aa36d866 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx2.c @@ -0,0 +1 @@ +#include "test-double-libmvec-sinh.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx512f.c b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx512f.c new file mode 100644 index 0000000000..55aa36d866 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-sinh-avx512f.c @@ -0,0 +1 @@ +#include "test-double-libmvec-sinh.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-sinh.c b/sysdeps/x86_64/fpu/test-double-libmvec-sinh.c new file mode 100644 index 0000000000..82dcaf745d --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-sinh.c @@ -0,0 +1,3 @@ +#define LIBMVEC_TYPE double +#define LIBMVEC_FUNC sinh +#include "test-vector-abi-arg1.h" diff --git a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c index 0222f9f5b8..db136cc901 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVbN2v_exp2) VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVbN2v_exp10) VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVbN2v_cosh) VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVbN2v_expm1) +VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVbN2v_sinh) #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 1aad9faf9c..5fc09ac8c0 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c @@ -38,6 +38,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVdN4v_exp2) VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVdN4v_exp10) VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVdN4v_cosh) VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVdN4v_expm1) +VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVdN4v_sinh) #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 e404bf899d..26ef7fb365 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVcN4v_exp2) VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVcN4v_exp10) VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVcN4v_cosh) VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVcN4v_expm1) +VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVcN4v_sinh) #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 2b4de59343..c7055fca76 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVeN8v_exp2) VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVeN8v_exp10) VECTOR_WRAPPER (WRAPPER_NAME (cosh), _ZGVeN8v_cosh) VECTOR_WRAPPER (WRAPPER_NAME (expm1), _ZGVeN8v_expm1) +VECTOR_WRAPPER (WRAPPER_NAME (sinh), _ZGVeN8v_sinh) #ifndef __ILP32__ # define VEC_INT_TYPE __m512i diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx.c b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx.c new file mode 100644 index 0000000000..93986945f3 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx.c @@ -0,0 +1 @@ +#include "test-float-libmvec-sinhf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx2.c b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx2.c new file mode 100644 index 0000000000..93986945f3 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx2.c @@ -0,0 +1 @@ +#include "test-float-libmvec-sinhf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx512f.c b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx512f.c new file mode 100644 index 0000000000..93986945f3 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf-avx512f.c @@ -0,0 +1 @@ +#include "test-float-libmvec-sinhf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-sinhf.c b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf.c new file mode 100644 index 0000000000..fb1f3c5c48 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-sinhf.c @@ -0,0 +1,3 @@ +#define LIBMVEC_TYPE float +#define LIBMVEC_FUNC sinhf +#include "test-vector-abi-arg1.h" diff --git a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c index 9a4a1b84a9..d353bcb0f2 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVeN16v_exp2f) VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVeN16v_exp10f) VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVeN16v_coshf) VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVeN16v_expm1f) +VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVeN16v_sinhf) #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 eb4e36d0e2..5e59117626 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVbN4v_exp2f) VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVbN4v_exp10f) VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVbN4v_coshf) VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVbN4v_expm1f) +VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVbN4v_sinhf) #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 d8adab59e6..e884a5f4df 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c @@ -38,6 +38,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVdN8v_exp2f) VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVdN8v_exp10f) VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVdN8v_coshf) VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVdN8v_expm1f) +VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVdN8v_sinhf) /* 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 e6e1a90c72..95910d39e9 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVcN8v_exp2f) VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVcN8v_exp10f) VECTOR_WRAPPER (WRAPPER_NAME (coshf), _ZGVcN8v_coshf) VECTOR_WRAPPER (WRAPPER_NAME (expm1f), _ZGVcN8v_expm1f) +VECTOR_WRAPPER (WRAPPER_NAME (sinhf), _ZGVcN8v_sinhf) #define VEC_INT_TYPE __m128i