1 | /* Single-precision vector (SVE) cos 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 neg_pio2_1, neg_pio2_2, neg_pio2_3, inv_pio2, shift; |
25 | } data = { |
26 | /* Polynomial coefficients are hard-wired in FTMAD instructions. */ |
27 | .neg_pio2_1 = -0x1.921fb6p+0f, |
28 | .neg_pio2_2 = 0x1.777a5cp-25f, |
29 | .neg_pio2_3 = 0x1.ee59dap-50f, |
30 | .inv_pio2 = 0x1.45f306p-1f, |
31 | /* Original shift used in AdvSIMD cosf, |
32 | plus a contribution to set the bit #0 of q |
33 | as expected by trigonometric instructions. */ |
34 | .shift = 0x1.800002p+23f |
35 | }; |
36 | |
37 | #define RangeVal 0x49800000 /* asuint32(0x1p20f). */ |
38 | |
39 | static svfloat32_t NOINLINE |
40 | special_case (svfloat32_t x, svfloat32_t y, svbool_t oob) |
41 | { |
42 | return sv_call_f32 (f: cosf, x, y, cmp: oob); |
43 | } |
44 | |
45 | /* A fast SVE implementation of cosf based on trigonometric |
46 | instructions (FTMAD, FTSSEL, FTSMUL). |
47 | Maximum measured error: 2.06 ULPs. |
48 | SV_NAME_F1 (cos)(0x1.dea2f2p+19) got 0x1.fffe7ap-6 |
49 | want 0x1.fffe76p-6. */ |
50 | svfloat32_t SV_NAME_F1 (cos) (svfloat32_t x, const svbool_t pg) |
51 | { |
52 | const struct data *d = ptr_barrier (&data); |
53 | |
54 | svfloat32_t r = svabs_x (pg, x); |
55 | svbool_t oob = svcmpge (pg, svreinterpret_u32 (r), RangeVal); |
56 | |
57 | /* Load some constants in quad-word chunks to minimise memory access. */ |
58 | svfloat32_t negpio2_and_invpio2 = svld1rq (svptrue_b32 (), &d->neg_pio2_1); |
59 | |
60 | /* n = rint(|x|/(pi/2)). */ |
61 | svfloat32_t q = svmla_lane (sv_f32 (x: d->shift), r, negpio2_and_invpio2, 3); |
62 | svfloat32_t n = svsub_x (pg, q, d->shift); |
63 | |
64 | /* r = |x| - n*(pi/2) (range reduction into -pi/4 .. pi/4). */ |
65 | r = svmla_lane (r, n, negpio2_and_invpio2, 0); |
66 | r = svmla_lane (r, n, negpio2_and_invpio2, 1); |
67 | r = svmla_lane (r, n, negpio2_and_invpio2, 2); |
68 | |
69 | /* Final multiplicative factor: 1.0 or x depending on bit #0 of q. */ |
70 | svfloat32_t f = svtssel (r, svreinterpret_u32 (q)); |
71 | |
72 | /* cos(r) poly approx. */ |
73 | svfloat32_t r2 = svtsmul (r, svreinterpret_u32 (q)); |
74 | svfloat32_t y = sv_f32 (x: 0.0f); |
75 | y = svtmad (y, r2, 4); |
76 | y = svtmad (y, r2, 3); |
77 | y = svtmad (y, r2, 2); |
78 | y = svtmad (y, r2, 1); |
79 | y = svtmad (y, r2, 0); |
80 | |
81 | if (__glibc_unlikely (svptest_any (pg, oob))) |
82 | return special_case (x, y: svmul_x (svnot_z (pg, oob), f, y), oob); |
83 | /* Apply factor. */ |
84 | return svmul_x (pg, f, y); |
85 | } |
86 | |