1 | //===-- Single-precision asinh function -----------------------------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | |
9 | #include "src/math/asinhf.h" |
10 | #include "src/__support/FPUtil/FPBits.h" |
11 | #include "src/__support/FPUtil/PolyEval.h" |
12 | #include "src/__support/FPUtil/multiply_add.h" |
13 | #include "src/__support/FPUtil/sqrt.h" |
14 | #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY |
15 | #include "src/math/generic/common_constants.h" |
16 | #include "src/math/generic/explogxf.h" |
17 | |
18 | namespace LIBC_NAMESPACE { |
19 | |
20 | LLVM_LIBC_FUNCTION(float, asinhf, (float x)) { |
21 | using FPBits_t = typename fputil::FPBits<float>; |
22 | FPBits_t xbits(x); |
23 | uint32_t x_u = xbits.uintval(); |
24 | uint32_t x_abs = xbits.abs().uintval(); |
25 | |
26 | // |x| <= 2^-3 |
27 | if (LIBC_UNLIKELY(x_abs <= 0x3e80'0000U)) { |
28 | // |x| <= 2^-26 |
29 | if (LIBC_UNLIKELY(x_abs <= 0x3280'0000U)) { |
30 | return static_cast<float>(LIBC_UNLIKELY(x_abs == 0) |
31 | ? x |
32 | : (x - 0x1.5555555555555p-3 * x * x * x)); |
33 | } |
34 | |
35 | double x_d = x; |
36 | double x_sq = x_d * x_d; |
37 | // Generated by Sollya with: |
38 | // > P = fpminimax(asinh(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16], [|D...|], |
39 | // [0, 2^-2]); |
40 | double p = fputil::polyeval( |
41 | x: x_sq, a0: 0.0, a: -0x1.555555555551ep-3, a: 0x1.3333333325495p-4, |
42 | a: -0x1.6db6db5a7622bp-5, a: 0x1.f1c70f82928c6p-6, a: -0x1.6e893934266b7p-6, |
43 | a: 0x1.1c0b41d3fbe78p-6, a: -0x1.c0f47810b3c4fp-7, a: 0x1.2c8602690143dp-7); |
44 | return static_cast<float>(fputil::multiply_add(x: x_d, y: p, z: x_d)); |
45 | } |
46 | |
47 | const double SIGN[2] = {1.0, -1.0}; |
48 | double x_sign = SIGN[x_u >> 31]; |
49 | double x_d = x; |
50 | |
51 | // Helper functions to set results for exceptional cases. |
52 | auto round_result_slightly_down = [x_sign](float r) -> float { |
53 | return fputil::multiply_add(x: static_cast<float>(x_sign), y: r, |
54 | z: static_cast<float>(x_sign) * (-0x1.0p-24f)); |
55 | }; |
56 | auto round_result_slightly_up = [x_sign](float r) -> float { |
57 | return fputil::multiply_add(x: static_cast<float>(x_sign), y: r, |
58 | z: static_cast<float>(x_sign) * 0x1.0p-24f); |
59 | }; |
60 | |
61 | if (LIBC_UNLIKELY(x_abs >= 0x4bdd'65a5U)) { |
62 | if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) |
63 | return x; |
64 | |
65 | // Exceptional cases when x > 2^24. |
66 | switch (x_abs) { |
67 | case 0x4bdd65a5: // |x| = 0x1.bacb4ap24f |
68 | return round_result_slightly_down(0x1.1e0696p4f); |
69 | case 0x4c803f2c: // |x| = 0x1.007e58p26f |
70 | return round_result_slightly_down(0x1.2b786cp4f); |
71 | case 0x4f8ffb03: // |x| = 0x1.1ff606p32f |
72 | return round_result_slightly_up(0x1.6fdd34p4f); |
73 | case 0x5c569e88: // |x| = 0x1.ad3d1p57f |
74 | return round_result_slightly_up(0x1.45c146p5f); |
75 | case 0x5e68984e: // |x| = 0x1.d1309cp61f |
76 | return round_result_slightly_up(0x1.5c9442p5f); |
77 | case 0x655890d3: // |x| = 0x1.b121a6p75f |
78 | return round_result_slightly_down(0x1.a9a3f2p5f); |
79 | case 0x65de7ca6: // |x| = 0x1.bcf94cp76f |
80 | return round_result_slightly_up(0x1.af66cp5f); |
81 | case 0x6eb1a8ec: // |x| = 0x1.6351d8p94f |
82 | return round_result_slightly_down(0x1.08b512p6f); |
83 | case 0x7997f30a: // |x| = 0x1.2fe614p116f |
84 | return round_result_slightly_up(0x1.451436p6f); |
85 | } |
86 | } else { |
87 | // Exceptional cases when x < 2^24. |
88 | if (LIBC_UNLIKELY(x_abs == 0x45abaf26)) { |
89 | // |x| = 0x1.575e4cp12f |
90 | return round_result_slightly_down(0x1.29becap3f); |
91 | } |
92 | if (LIBC_UNLIKELY(x_abs == 0x49d29048)) { |
93 | // |x| = 0x1.a5209p20f |
94 | return round_result_slightly_down(0x1.e1b92p3f); |
95 | } |
96 | } |
97 | |
98 | // asinh(x) = log(x + sqrt(x^2 + 1)) |
99 | return static_cast<float>( |
100 | x_sign * |
101 | log_eval(x: fputil::multiply_add( |
102 | x: x_d, y: x_sign, z: fputil::sqrt(x: fputil::multiply_add(x: x_d, y: x_d, z: 1.0))))); |
103 | } |
104 | |
105 | } // namespace LIBC_NAMESPACE |
106 | |