| 1 | //===-- Implementation header for asinhf16 ----------------------*- 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_ASINHF16_H |
| 10 | #define LLVM_LIBC_SRC___SUPPORT_MATH_ASINHF16_H |
| 11 | |
| 12 | #include "include/llvm-libc-macros/float16-macros.h" |
| 13 | |
| 14 | #ifdef LIBC_TYPES_HAS_FLOAT16 |
| 15 | |
| 16 | #include "acoshf_utils.h" |
| 17 | #include "src/__support/FPUtil/FEnvImpl.h" |
| 18 | #include "src/__support/FPUtil/FPBits.h" |
| 19 | #include "src/__support/FPUtil/PolyEval.h" |
| 20 | #include "src/__support/FPUtil/cast.h" |
| 21 | #include "src/__support/FPUtil/except_value_utils.h" |
| 22 | #include "src/__support/FPUtil/multiply_add.h" |
| 23 | #include "src/__support/FPUtil/rounding_mode.h" |
| 24 | #include "src/__support/FPUtil/sqrt.h" |
| 25 | #include "src/__support/macros/config.h" |
| 26 | #include "src/__support/macros/optimization.h" |
| 27 | |
| 28 | namespace LIBC_NAMESPACE_DECL { |
| 29 | |
| 30 | namespace math { |
| 31 | |
| 32 | LIBC_INLINE constexpr float16 asinhf16(float16 x) { |
| 33 | |
| 34 | #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 35 | constexpr size_t N_EXCEPTS = 8; |
| 36 | |
| 37 | constexpr fputil::ExceptValues<float16, N_EXCEPTS> ASINHF16_EXCEPTS{.values: { |
| 38 | // (input, RZ output, RU offset, RD offset, RN offset) |
| 39 | |
| 40 | // x = 0x1.da4p-2, asinhf16(x) = 0x1.ca8p-2 (RZ) |
| 41 | {.input: 0x3769, .rnd_towardzero_result: 0x372a, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 42 | // x = 0x1.d6cp-1, asinhf16(x) = 0x1.a58p-1 (RZ) |
| 43 | {.input: 0x3b5b, .rnd_towardzero_result: 0x3a96, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 0}, |
| 44 | // x = 0x1.c7cp+3, asinhf16(x) = 0x1.accp+1 (RZ) |
| 45 | {.input: 0x4b1f, .rnd_towardzero_result: 0x42b3, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 0}, |
| 46 | // x = 0x1.26cp+4, asinhf16(x) = 0x1.cd8p+1 (RZ) |
| 47 | {.input: 0x4c9b, .rnd_towardzero_result: 0x4336, .rnd_upward_offset: 1, .rnd_downward_offset: 0, .rnd_tonearest_offset: 1}, |
| 48 | // x = -0x1.da4p-2, asinhf16(x) = -0x1.ca8p-2 (RZ) |
| 49 | {.input: 0xb769, .rnd_towardzero_result: 0xb72a, .rnd_upward_offset: 0, .rnd_downward_offset: 1, .rnd_tonearest_offset: 1}, |
| 50 | // x = -0x1.d6cp-1, asinhf16(x) = -0x1.a58p-1 (RZ) |
| 51 | {.input: 0xbb5b, .rnd_towardzero_result: 0xba96, .rnd_upward_offset: 0, .rnd_downward_offset: 1, .rnd_tonearest_offset: 0}, |
| 52 | // x = -0x1.c7cp+3, asinhf16(x) = -0x1.accp+1 (RZ) |
| 53 | {.input: 0xcb1f, .rnd_towardzero_result: 0xc2b3, .rnd_upward_offset: 0, .rnd_downward_offset: 1, .rnd_tonearest_offset: 0}, |
| 54 | // x = -0x1.26cp+4, asinhf16(x) = -0x1.cd8p+1 (RZ) |
| 55 | {.input: 0xcc9b, .rnd_towardzero_result: 0xc336, .rnd_upward_offset: 0, .rnd_downward_offset: 1, .rnd_tonearest_offset: 1}, |
| 56 | }}; |
| 57 | #endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 58 | |
| 59 | using namespace acoshf_internal; |
| 60 | using FPBits = fputil::FPBits<float16>; |
| 61 | FPBits xbits(x); |
| 62 | |
| 63 | uint16_t x_u = xbits.uintval(); |
| 64 | uint16_t x_abs = x_u & 0x7fff; |
| 65 | |
| 66 | if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) { |
| 67 | if (xbits.is_signaling_nan()) { |
| 68 | fputil::raise_except_if_required(FE_INVALID); |
| 69 | return FPBits::quiet_nan().get_val(); |
| 70 | } |
| 71 | |
| 72 | return x; |
| 73 | } |
| 74 | |
| 75 | #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 76 | // Handle exceptional values |
| 77 | if (auto r = ASINHF16_EXCEPTS.lookup(x_bits: x_u); LIBC_UNLIKELY(r.has_value())) |
| 78 | return r.value(); |
| 79 | #endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS |
| 80 | |
| 81 | float xf = x; |
| 82 | const float SIGN[2] = {1.0f, -1.0f}; |
| 83 | float x_sign = SIGN[x_u >> 15]; |
| 84 | |
| 85 | // |x| <= 0.25 |
| 86 | if (LIBC_UNLIKELY(x_abs <= 0x3400)) { |
| 87 | // when |x| < 0x1.718p-5, asinhf16(x) = x. Adjust by 1 ULP for certain |
| 88 | // rounding types. |
| 89 | if (LIBC_UNLIKELY(x_abs < 0x29c6)) { |
| 90 | #ifndef LIBC_MATH_HAS_ASSUME_ROUND_NEAREST_ONLY |
| 91 | int rounding = fputil::quick_get_round(); |
| 92 | if ((rounding == FE_UPWARD || rounding == FE_TOWARDZERO) && xf < 0) |
| 93 | return fputil::cast<float16>(x: xf + 0x1p-24f); |
| 94 | if ((rounding == FE_DOWNWARD || rounding == FE_TOWARDZERO) && xf > 0) |
| 95 | return fputil::cast<float16>(x: xf - 0x1p-24f); |
| 96 | #endif |
| 97 | return fputil::cast<float16>(x: xf); |
| 98 | } |
| 99 | |
| 100 | float x_sq = xf * xf; |
| 101 | // Generated by Sollya with: |
| 102 | // > P = fpminimax(asinh(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 2^-2]); |
| 103 | // The last coefficient 0x1.bd114ep-6f has been changed to 0x1.bd114ep-5f |
| 104 | // for better accuracy. |
| 105 | float p = fputil::polyeval(x: x_sq, a0: 1.0f, a: -0x1.555552p-3f, a: 0x1.332f6ap-4f, |
| 106 | a: -0x1.6c53dep-5f, a: 0x1.bd114ep-5f); |
| 107 | |
| 108 | return fputil::cast<float16>(x: xf * p); |
| 109 | } |
| 110 | |
| 111 | // General case: asinh(x) = ln(x + sqrt(x^2 + 1)) |
| 112 | float sqrt_term = fputil::sqrt<float>(x: fputil::multiply_add(x: xf, y: xf, z: 1.0f)); |
| 113 | return fputil::cast<float16>( |
| 114 | x: x_sign * log_eval(x: fputil::multiply_add(x: xf, y: x_sign, z: sqrt_term))); |
| 115 | } |
| 116 | |
| 117 | } // namespace math |
| 118 | |
| 119 | } // namespace LIBC_NAMESPACE_DECL |
| 120 | |
| 121 | #endif // LIBC_TYPES_HAS_FLOAT16 |
| 122 | |
| 123 | #endif // LLVM_LIBC_SRC___SUPPORT_MATH_ASINHF16_H |
| 124 | |