| 1 | //===-- Implementation header for atanf16 -----------------------*- C++ -*-===// |
| 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 | #ifndef LLVM_LIBC_SRC___SUPPORT_MATH_ATANF16_H |
| 10 | #define LLVM_LIBC_SRC___SUPPORT_MATH_ATANF16_H |
| 11 | |
| 12 | #include "include/llvm-libc-macros/float16-macros.h" |
| 13 | |
| 14 | #ifdef LIBC_TYPES_HAS_FLOAT16 |
| 15 | |
| 16 | #include "src/__support/FPUtil/FEnvImpl.h" |
| 17 | #include "src/__support/FPUtil/FPBits.h" |
| 18 | #include "src/__support/FPUtil/PolyEval.h" |
| 19 | #include "src/__support/FPUtil/cast.h" |
| 20 | #include "src/__support/FPUtil/except_value_utils.h" |
| 21 | #include "src/__support/FPUtil/multiply_add.h" |
| 22 | #include "src/__support/FPUtil/sqrt.h" |
| 23 | #include "src/__support/macros/optimization.h" |
| 24 | |
| 25 | namespace LIBC_NAMESPACE_DECL { |
| 26 | |
| 27 | namespace math { |
| 28 | |
| 29 | LIBC_INLINE constexpr float16 atanf16(float16 x) { |
| 30 | // Generated by Solly using the following command: |
| 31 | // > round(pi/2, SG, RN); |
| 32 | constexpr float PI_2 = 0x1.921fb6p0; |
| 33 | |
| 34 | #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 35 | constexpr size_t N_EXCEPTS = 6; |
| 36 | |
| 37 | constexpr fputil::ExceptValues<float16, N_EXCEPTS> ATANF16_EXCEPTS{.values: { |
| 38 | // (input, RZ output, RU offset, RD offset, RN offset) |
| 39 | {.input: 0x2745, .rnd_towardzero_result: 0x2744, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 40 | {.input: 0x3099, .rnd_towardzero_result: 0x3090, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 41 | {.input: 0x3c6c, .rnd_towardzero_result: 0x3aae, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 42 | {.input: 0x466e, .rnd_towardzero_result: 0x3daa, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 43 | {.input: 0x48ae, .rnd_towardzero_result: 0x3ddb, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 0}, |
| 44 | {.input: 0x5619, .rnd_towardzero_result: 0x3e3d, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 45 | }}; |
| 46 | #endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 47 | |
| 48 | using FPBits = fputil::FPBits<float16>; |
| 49 | FPBits xbits(x); |
| 50 | |
| 51 | uint16_t x_u = xbits.uintval(); |
| 52 | uint16_t x_abs = x_u & 0x7fff; |
| 53 | bool x_sign = x_u >> 15; |
| 54 | float sign = (x_sign ? -1.0 : 1.0); |
| 55 | |
| 56 | // |x| >= +/-inf |
| 57 | if (LIBC_UNLIKELY(x_abs >= 0x7c00)) { |
| 58 | if (xbits.is_nan()) { |
| 59 | if (xbits.is_signaling_nan()) { |
| 60 | fputil::raise_except_if_required(FE_INVALID); |
| 61 | return FPBits::quiet_nan().get_val(); |
| 62 | } |
| 63 | return x; |
| 64 | } |
| 65 | |
| 66 | // atanf16(+/-inf) = +/-pi/2 |
| 67 | return fputil::cast<float16>(x: sign * PI_2); |
| 68 | } |
| 69 | |
| 70 | float xf = x; |
| 71 | float xsq = xf * xf; |
| 72 | #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 73 | // Handle exceptional values |
| 74 | if (auto r = ATANF16_EXCEPTS.lookup_odd(x_abs, sign: x_sign); |
| 75 | LIBC_UNLIKELY(r.has_value())) |
| 76 | return r.value(); |
| 77 | #endif |
| 78 | |
| 79 | // |x| <= 0x1p0, |x| <= 1 |
| 80 | if (x_abs <= 0x3c00) { |
| 81 | // atanf16(+/-0) = +/-0 |
| 82 | if (LIBC_UNLIKELY(x_abs == 0)) |
| 83 | return x; |
| 84 | |
| 85 | // Degree-14 minimax odd polynomial of atan(x) generated by Sollya with: |
| 86 | // > P = fpminimax(atan(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14|], [|SG...|], |
| 87 | // [0, 1]); |
| 88 | float result = fputil::polyeval( |
| 89 | x: xsq, a0: 0x1.fffffcp-1f, a: -0x1.55519ep-2f, a: 0x1.98f6a8p-3f, a: -0x1.1f0a92p-3f, |
| 90 | a: 0x1.95b654p-4f, a: -0x1.e65492p-5f, a: 0x1.8c0c36p-6f, a: -0x1.32316ep-8f); |
| 91 | return fputil::cast<float16>(x: xf * result); |
| 92 | } |
| 93 | |
| 94 | // If |x| > 1 |
| 95 | // y = atan(x) = sign(x) * atan(|x|) |
| 96 | // atan(|x|) = pi/2 - atan(1/|x|) |
| 97 | // Recall, 1/|x| < 1 |
| 98 | float x_inv_sq = 1.0f / xsq; |
| 99 | float x_inv = fputil::sqrt<float>(x: x_inv_sq); |
| 100 | |
| 101 | // Degree-14 minimax odd polynomial of atan(x) generated by Sollya with: |
| 102 | // > P = fpminimax(atan(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14|], [|SG...|], |
| 103 | // [0, 1]); |
| 104 | float interm = |
| 105 | fputil::polyeval(x: x_inv_sq, a0: 0x1.fffffcp-1f, a: -0x1.55519ep-2f, |
| 106 | a: 0x1.98f6a8p-3f, a: -0x1.1f0a92p-3f, a: 0x1.95b654p-4f, |
| 107 | a: -0x1.e65492p-5f, a: 0x1.8c0c36p-6f, a: -0x1.32316ep-8f); |
| 108 | |
| 109 | return fputil::cast<float16>(x: sign * |
| 110 | fputil::multiply_add(x: x_inv, y: -interm, z: PI_2)); |
| 111 | } |
| 112 | |
| 113 | } // namespace math |
| 114 | |
| 115 | } // namespace LIBC_NAMESPACE_DECL |
| 116 | |
| 117 | #endif // LIBC_TYPES_HAS_FLOAT16 |
| 118 | |
| 119 | #endif // LLVM_LIBC_SRC___SUPPORT_MATH_ATANF16_H |
| 120 | |