From: Adhemerval Zanella Date: Tue, 19 Nov 2024 19:30:16 +0000 (-0300) Subject: math: Use tanhf from CORE-MATH X-Git-Tag: glibc-2.41~294 X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=0e0be3ed803714285831b14a0032bc91b93e9027;p=thirdparty%2Fglibc.git math: Use tanhf from CORE-MATH The CORE-MATH implementation is correctly rounded (for any rounding mode) and shows slight better performance to the generic tanhf. The code was adapted to glibc style and to use the definition of math_config.h (to handle errno, overflow, and underflow). Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1, gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1): Latency master patched improvement x86_64 51.5273 41.0951 20.25% x86_64v2 47.7021 39.1526 17.92% x86_64v3 45.0373 34.2737 23.90% i686 133.9970 83.8596 37.42% aarch64 (Neoverse) 21.5439 14.7961 31.32% power10 13.3301 8.4406 36.68% reciprocal-throughput master patched improvement x86_64 24.9493 12.8547 48.48% x86_64v2 20.7051 12.7761 38.29% x86_64v3 19.2492 11.0851 42.41% i686 78.6498 29.8211 62.08% aarch64 (Neoverse) 11.6026 7.11487 38.68% power10 6.3328 2.8746 54.61% Signed-off-by: Alexei Sibidanov Signed-off-by: Paul Zimmermann Signed-off-by: Adhemerval Zanella Reviewed-by: DJ Delorie --- diff --git a/SHARED-FILES b/SHARED-FILES index 3bd4e7fb4a..032c407881 100644 --- a/SHARED-FILES +++ b/SHARED-FILES @@ -330,3 +330,7 @@ sysdeps/ieee754/flt-32/e_sinhf.c: (src/binary32/sinh/sinhf.c in CORE-MATH) - the code was adapted to use glibc code style and internal functions to handle errno, overflow, and underflow. +sysdeps/ieee754/flt-32/s_tanhf.c: + (src/binary32/tanh/tanhf.c in CORE-MATH) + - the code was adapted to use glibc code style and internal + functions to handle errno, overflow, and underflow. diff --git a/sysdeps/aarch64/libm-test-ulps b/sysdeps/aarch64/libm-test-ulps index 1e9c5af0ca..190685f893 100644 --- a/sysdeps/aarch64/libm-test-ulps +++ b/sysdeps/aarch64/libm-test-ulps @@ -1665,7 +1665,6 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_advsimd": @@ -1674,7 +1673,6 @@ float: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_sve": @@ -1683,12 +1681,10 @@ float: 2 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tanpi": diff --git a/sysdeps/alpha/fpu/libm-test-ulps b/sysdeps/alpha/fpu/libm-test-ulps index 7d5fcc7f2e..ddc851b73c 100644 --- a/sysdeps/alpha/fpu/libm-test-ulps +++ b/sysdeps/alpha/fpu/libm-test-ulps @@ -1318,22 +1318,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tgamma": diff --git a/sysdeps/arc/fpu/libm-test-ulps b/sysdeps/arc/fpu/libm-test-ulps index ec42c5eac9..92fad18fc5 100644 --- a/sysdeps/arc/fpu/libm-test-ulps +++ b/sysdeps/arc/fpu/libm-test-ulps @@ -1053,19 +1053,15 @@ double: 1 Function: "tanh": double: 3 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 3 -float: 3 Function: "tanh_upward": double: 3 -float: 3 Function: "tgamma": double: 9 diff --git a/sysdeps/arc/nofpu/libm-test-ulps b/sysdeps/arc/nofpu/libm-test-ulps index 15b2b82187..6fb7ce93f4 100644 --- a/sysdeps/arc/nofpu/libm-test-ulps +++ b/sysdeps/arc/nofpu/libm-test-ulps @@ -251,7 +251,6 @@ double: 2 Function: "tanh": double: 2 -float: 2 Function: "tgamma": double: 9 diff --git a/sysdeps/arm/libm-test-ulps b/sysdeps/arm/libm-test-ulps index 201099d3f2..3ff089c1eb 100644 --- a/sysdeps/arm/libm-test-ulps +++ b/sysdeps/arm/libm-test-ulps @@ -1143,19 +1143,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tanpi": double: 2 diff --git a/sysdeps/csky/fpu/libm-test-ulps b/sysdeps/csky/fpu/libm-test-ulps index 3824f00415..6ba3cef623 100644 --- a/sysdeps/csky/fpu/libm-test-ulps +++ b/sysdeps/csky/fpu/libm-test-ulps @@ -971,19 +971,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tgamma": double: 9 diff --git a/sysdeps/csky/nofpu/libm-test-ulps b/sysdeps/csky/nofpu/libm-test-ulps index 34f5eb52a2..4f428a65c5 100644 --- a/sysdeps/csky/nofpu/libm-test-ulps +++ b/sysdeps/csky/nofpu/libm-test-ulps @@ -1002,19 +1002,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tgamma": double: 9 diff --git a/sysdeps/hppa/fpu/libm-test-ulps b/sysdeps/hppa/fpu/libm-test-ulps index e48fb8a66e..759e224308 100644 --- a/sysdeps/hppa/fpu/libm-test-ulps +++ b/sysdeps/hppa/fpu/libm-test-ulps @@ -1176,19 +1176,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tanpi": double: 2 diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps index d545682fd9..423061e12e 100644 --- a/sysdeps/i386/fpu/libm-test-ulps +++ b/sysdeps/i386/fpu/libm-test-ulps @@ -1754,25 +1754,21 @@ ldouble: 2 Function: "tanh": double: 2 -float: 2 float128: 2 ldouble: 3 Function: "tanh_downward": double: 3 -float: 3 float128: 4 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 float128: 3 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 float128: 3 ldouble: 4 diff --git a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps index ad9a52749b..466e68dae0 100644 --- a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps +++ b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps @@ -1759,25 +1759,21 @@ ldouble: 2 Function: "tanh": double: 2 -float: 2 float128: 2 ldouble: 3 Function: "tanh_downward": double: 3 -float: 3 float128: 4 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 float128: 3 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 float128: 3 ldouble: 4 diff --git a/sysdeps/ieee754/flt-32/s_tanhf.c b/sysdeps/ieee754/flt-32/s_tanhf.c index 2c12f04569..e07f3589b8 100644 --- a/sysdeps/ieee754/flt-32/s_tanhf.c +++ b/sysdeps/ieee754/flt-32/s_tanhf.c @@ -1,63 +1,90 @@ -/* s_tanhf.c -- float version of s_tanh.c. - */ +/* Correctly-rounded hyperbolic tangent function for binary32 value. -/* - * ==================================================== - * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. - * - * Developed at SunPro, a Sun Microsystems, Inc. business. - * Permission to use, copy, modify, and distribute this - * software is freely granted, provided that this notice - * is preserved. - * ==================================================== - */ +Copyright (c) 2022-2024 Alexei Sibidanov. -#if defined(LIBM_SCCS) && !defined(lint) -static char rcsid[] = "$NetBSD: s_tanhf.c,v 1.4 1995/05/10 20:48:24 jtc Exp $"; -#endif +The original version of this file was copied from the CORE-MATH +project (file src/binary32/tanh/tanhf.c, revision b1ecd83). -#include -#include -#include -#include -#include +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: -static const float one=1.0, two=2.0, tiny = 1.0e-30; - -float __tanhf(float x) -{ - float t,z; - int32_t jx,ix; +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. - GET_FLOAT_WORD(jx,x); - ix = jx&0x7fffffff; +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +*/ - /* x is INF or NaN */ - if(ix>=0x7f800000) { - if (jx>=0) return one/x+one; /* tanh(+-inf)=+-1 */ - else return one/x-one; /* tanh(NaN) = NaN */ - } +#include +#include +#include +#include "math_config.h" - /* |x| < 22 */ - if (ix < 0x41b00000) { /* |x|<22 */ - if (ix == 0) - return x; /* x == +-0 */ - if (ix<0x24000000) /* |x|<2**-55 */ - { - math_check_force_underflow (x); - return x*(one+x); /* tanh(small) = small */ - } - if (ix>=0x3f800000) { /* |x|>=1 */ - t = __expm1f(two*fabsf(x)); - z = one - two/(t+two); - } else { - t = __expm1f(-two*fabsf(x)); - z= -t/(t+two); - } - /* |x| > 22, return +-1 */ - } else { - z = one - tiny; /* raised inexact flag */ +float +__tanhf (float x) +{ + double z = x; + uint32_t ux = asuint (x); + int e = (ux >> 23) & 0xff; + if (__glibc_unlikely (e == 0xff)) + { + if (ux << 9) + return x + x; /* x = nan */ + static const float ir[] = { 1.0f, -1.0f }; + return ir[ux >> 31]; /* x = +-inf */ + } + if (__glibc_unlikely (e < 115)) + { + if (__glibc_unlikely (e < 102)) + { + if (__glibc_unlikely ((ux << 1) == 0)) + return x; + return fmaf (-x, fabsf (x), x); } - return (jx>=0)? z: -z; + float x2 = x * x; + return fmaf (x, -0x1.555556p-2f * x2, x); + } + if ((ux << 1) > (0x41102cb3u << 1)) + return copysignf (1.0f, x) - copysignf (0x1p-25f, x); + double z2 = z * z; + double z4 = z2 * z2; + double z8 = z4 * z4; + static const double cn[] = + { + 0x1p+0, 0x1.30877b8b72d33p-3, 0x1.694aa09ae9e5ep-8, + 0x1.4101377abb729p-14, 0x1.e0392b1db0018p-22, 0x1.2533756e546f7p-30, + 0x1.d62e5abe6ae8ap-41, 0x1.b06be534182dep-54 + }; + static const double cd[] = + { + 0x1p+0, 0x1.ed99131b0ebeap-2, 0x1.0d27ed6c95a69p-5, + 0x1.7cbdaca0e9fccp-11, 0x1.b4e60b892578ep-18, 0x1.a6f707c5c71abp-26, + 0x1.35a8b6e2cd94cp-35, 0x1.ca8230677aa01p-47 + }; + double n0 = cn[0] + z2 * cn[1]; + double n2 = cn[2] + z2 * cn[3]; + double n4 = cn[4] + z2 * cn[5]; + double n6 = cn[6] + z2 * cn[7]; + n0 += z4 * n2; + n4 += z4 * n6; + n0 += z8 * n4; + double d0 = cd[0] + z2 * cd[1]; + double d2 = cd[2] + z2 * cd[3]; + double d4 = cd[4] + z2 * cd[5]; + double d6 = cd[6] + z2 * cd[7]; + d0 += z4 * d2; + d4 += z4 * d6; + d0 += z8 * d4; + double r = z * n0 / d0; + return r; } libm_alias_float (__tanh, tanh) diff --git a/sysdeps/loongarch/lp64/libm-test-ulps b/sysdeps/loongarch/lp64/libm-test-ulps index 1e57c6a4df..0840e9e190 100644 --- a/sysdeps/loongarch/lp64/libm-test-ulps +++ b/sysdeps/loongarch/lp64/libm-test-ulps @@ -1325,22 +1325,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tgamma": diff --git a/sysdeps/m68k/m680x0/fpu/libm-test-ulps b/sysdeps/m68k/m680x0/fpu/libm-test-ulps index 1ef5e0e5a3..6acb5cd043 100644 --- a/sysdeps/m68k/m680x0/fpu/libm-test-ulps +++ b/sysdeps/m68k/m680x0/fpu/libm-test-ulps @@ -1152,15 +1152,12 @@ double: 1 Function: "tanh_downward": double: 1 -float: 1 Function: "tanh_towardzero": double: 1 -float: 1 Function: "tanh_upward": double: 1 -float: 1 Function: "tgamma": double: 3 diff --git a/sysdeps/microblaze/libm-test-ulps b/sysdeps/microblaze/libm-test-ulps index 0f0e1fb35d..7a6d75f91c 100644 --- a/sysdeps/microblaze/libm-test-ulps +++ b/sysdeps/microblaze/libm-test-ulps @@ -233,7 +233,6 @@ double: 2 Function: "tanh": double: 2 -float: 2 Function: "tgamma": double: 5 diff --git a/sysdeps/mips/mips32/libm-test-ulps b/sysdeps/mips/mips32/libm-test-ulps index 58223df3b4..3ede885f9e 100644 --- a/sysdeps/mips/mips32/libm-test-ulps +++ b/sysdeps/mips/mips32/libm-test-ulps @@ -1050,19 +1050,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tgamma": double: 9 diff --git a/sysdeps/mips/mips64/libm-test-ulps b/sysdeps/mips/mips64/libm-test-ulps index efa76ec249..891f887030 100644 --- a/sysdeps/mips/mips64/libm-test-ulps +++ b/sysdeps/mips/mips64/libm-test-ulps @@ -1336,22 +1336,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tgamma": diff --git a/sysdeps/or1k/fpu/libm-test-ulps b/sysdeps/or1k/fpu/libm-test-ulps index 5900e6d6d9..c27afc99c8 100644 --- a/sysdeps/or1k/fpu/libm-test-ulps +++ b/sysdeps/or1k/fpu/libm-test-ulps @@ -984,19 +984,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tgamma": double: 9 diff --git a/sysdeps/or1k/nofpu/libm-test-ulps b/sysdeps/or1k/nofpu/libm-test-ulps index 6f60e83486..46ef288135 100644 --- a/sysdeps/or1k/nofpu/libm-test-ulps +++ b/sysdeps/or1k/nofpu/libm-test-ulps @@ -1013,19 +1013,15 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_downward": double: 3 -float: 3 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tanh_upward": double: 3 -float: 3 Function: "tanpi": double: 2 diff --git a/sysdeps/powerpc/fpu/libm-test-ulps b/sysdeps/powerpc/fpu/libm-test-ulps index 9fe6f82a41..07b91f8534 100644 --- a/sysdeps/powerpc/fpu/libm-test-ulps +++ b/sysdeps/powerpc/fpu/libm-test-ulps @@ -1861,25 +1861,21 @@ ldouble: 3 Function: "tanh": double: 2 -float: 2 float128: 2 ldouble: 1 Function: "tanh_downward": double: 3 -float: 3 float128: 4 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 float128: 3 ldouble: 4 Function: "tanh_upward": double: 3 -float: 3 float128: 3 ldouble: 6 diff --git a/sysdeps/powerpc/nofpu/libm-test-ulps b/sysdeps/powerpc/nofpu/libm-test-ulps index 769c14e81d..3f02dc0be2 100644 --- a/sysdeps/powerpc/nofpu/libm-test-ulps +++ b/sysdeps/powerpc/nofpu/libm-test-ulps @@ -1452,22 +1452,18 @@ ldouble: 3 Function: "tanh": double: 2 -float: 2 ldouble: 1 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 4 Function: "tanh_upward": double: 3 -float: 3 ldouble: 6 Function: "tgamma": diff --git a/sysdeps/riscv/nofpu/libm-test-ulps b/sysdeps/riscv/nofpu/libm-test-ulps index 3ce20f4bb1..055826625c 100644 --- a/sysdeps/riscv/nofpu/libm-test-ulps +++ b/sysdeps/riscv/nofpu/libm-test-ulps @@ -1265,22 +1265,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tgamma": diff --git a/sysdeps/riscv/rvd/libm-test-ulps b/sysdeps/riscv/rvd/libm-test-ulps index 98146886d9..69e0eb2e42 100644 --- a/sysdeps/riscv/rvd/libm-test-ulps +++ b/sysdeps/riscv/rvd/libm-test-ulps @@ -1323,22 +1323,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tgamma": diff --git a/sysdeps/s390/fpu/libm-test-ulps b/sysdeps/s390/fpu/libm-test-ulps index e8a3754e4c..1a91bffcfe 100644 --- a/sysdeps/s390/fpu/libm-test-ulps +++ b/sysdeps/s390/fpu/libm-test-ulps @@ -1362,22 +1362,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tanpi": diff --git a/sysdeps/sh/libm-test-ulps b/sysdeps/sh/libm-test-ulps index 3c07cd1317..9378516d58 100644 --- a/sysdeps/sh/libm-test-ulps +++ b/sysdeps/sh/libm-test-ulps @@ -486,11 +486,9 @@ double: 1 Function: "tanh": double: 2 -float: 2 Function: "tanh_towardzero": double: 2 -float: 2 Function: "tgamma": double: 9 diff --git a/sysdeps/sparc/fpu/libm-test-ulps b/sysdeps/sparc/fpu/libm-test-ulps index ec9d2d9d21..9123fff50b 100644 --- a/sysdeps/sparc/fpu/libm-test-ulps +++ b/sysdeps/sparc/fpu/libm-test-ulps @@ -1377,22 +1377,18 @@ ldouble: 1 Function: "tanh": double: 2 -float: 2 ldouble: 2 Function: "tanh_downward": double: 3 -float: 3 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 ldouble: 3 Function: "tanpi": diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps index 7ce0714683..ad936ad32f 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -2280,25 +2280,21 @@ float: 2 Function: "tanh": double: 2 -float: 2 float128: 2 ldouble: 3 Function: "tanh_downward": double: 3 -float: 3 float128: 4 ldouble: 4 Function: "tanh_towardzero": double: 2 -float: 2 float128: 3 ldouble: 3 Function: "tanh_upward": double: 3 -float: 3 float128: 3 ldouble: 4