1 | /* Single-precision vector (SVE) sin function. |
2 | |
3 | Copyright (C) 2023-2024 Free Software Foundation, Inc. |
4 | This file is part of the GNU C Library. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2.1 of the License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; if not, see |
18 | <https://www.gnu.org/licenses/>. */ |
19 | |
20 | #include "sv_math.h" |
21 | |
22 | static const struct data |
23 | { |
24 | float poly[4]; |
25 | /* Pi-related values to be loaded as one quad-word and used with |
26 | svmla_lane. */ |
27 | float negpi1, negpi2, negpi3, invpi; |
28 | float shift; |
29 | } data = { |
30 | .poly = { |
31 | /* Non-zero coefficients from the degree 9 Taylor series expansion of |
32 | sin. */ |
33 | -0x1.555548p-3f, 0x1.110df4p-7f, -0x1.9f42eap-13f, 0x1.5b2e76p-19f |
34 | }, |
35 | .negpi1 = -0x1.921fb6p+1f, |
36 | .negpi2 = 0x1.777a5cp-24f, |
37 | .negpi3 = 0x1.ee59dap-49f, |
38 | .invpi = 0x1.45f306p-2f, |
39 | .shift = 0x1.8p+23f |
40 | }; |
41 | |
42 | #define RangeVal 0x49800000 /* asuint32 (0x1p20f). */ |
43 | #define C(i) sv_f32 (d->poly[i]) |
44 | |
45 | static svfloat32_t NOINLINE |
46 | special_case (svfloat32_t x, svfloat32_t y, svbool_t cmp) |
47 | { |
48 | return sv_call_f32 (f: sinf, x, y, cmp); |
49 | } |
50 | |
51 | /* A fast SVE implementation of sinf. |
52 | Maximum error: 1.89 ULPs. |
53 | This maximum error is achieved at multiple values in [-2^18, 2^18] |
54 | but one example is: |
55 | SV_NAME_F1 (sin)(0x1.9247a4p+0) got 0x1.fffff6p-1 want 0x1.fffffap-1. */ |
56 | svfloat32_t SV_NAME_F1 (sin) (svfloat32_t x, const svbool_t pg) |
57 | { |
58 | const struct data *d = ptr_barrier (&data); |
59 | |
60 | svfloat32_t ax = svabs_x (pg, x); |
61 | svuint32_t sign |
62 | = sveor_x (pg, svreinterpret_u32 (x), svreinterpret_u32 (ax)); |
63 | svbool_t cmp = svcmpge (pg, svreinterpret_u32 (ax), RangeVal); |
64 | |
65 | /* pi_vals are a quad-word of helper values - the first 3 elements contain |
66 | -pi in extended precision, the last contains 1 / pi. */ |
67 | svfloat32_t pi_vals = svld1rq (svptrue_b32 (), &d->negpi1); |
68 | |
69 | /* n = rint(|x|/pi). */ |
70 | svfloat32_t n = svmla_lane (sv_f32 (x: d->shift), ax, pi_vals, 3); |
71 | svuint32_t odd = svlsl_x (pg, svreinterpret_u32 (n), 31); |
72 | n = svsub_x (pg, n, d->shift); |
73 | |
74 | /* r = |x| - n*pi (range reduction into -pi/2 .. pi/2). */ |
75 | svfloat32_t r; |
76 | r = svmla_lane (ax, n, pi_vals, 0); |
77 | r = svmla_lane (r, n, pi_vals, 1); |
78 | r = svmla_lane (r, n, pi_vals, 2); |
79 | |
80 | /* sin(r) approx using a degree 9 polynomial from the Taylor series |
81 | expansion. Note that only the odd terms of this are non-zero. */ |
82 | svfloat32_t r2 = svmul_x (pg, r, r); |
83 | svfloat32_t y; |
84 | y = svmla_x (pg, C (2), r2, C (3)); |
85 | y = svmla_x (pg, C (1), r2, y); |
86 | y = svmla_x (pg, C (0), r2, y); |
87 | y = svmla_x (pg, r, r, svmul_x (pg, y, r2)); |
88 | |
89 | /* sign = y^sign^odd. */ |
90 | sign = sveor_x (pg, sign, odd); |
91 | |
92 | if (__glibc_unlikely (svptest_any (pg, cmp))) |
93 | return special_case (x, |
94 | y: svreinterpret_f32 (sveor_x ( |
95 | svnot_z (pg, cmp), svreinterpret_u32 (y), sign)), |
96 | cmp); |
97 | return svreinterpret_f32 (sveor_x (pg, svreinterpret_u32 (y), sign)); |
98 | } |
99 | |