brintos

brintos / llvm-project-archived public Read only

0
0
Text · 5.6 KiB · f6b5c9b Raw
154 lines · cpp
1//===-- Half-precision sinh(x) 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-exception6//7//===----------------------------------------------------------------------===//8 9#include "src/math/sinhf16.h"10#include "hdr/errno_macros.h"11#include "hdr/fenv_macros.h"12#include "src/__support/FPUtil/FEnvImpl.h"13#include "src/__support/FPUtil/FPBits.h"14#include "src/__support/FPUtil/except_value_utils.h"15#include "src/__support/FPUtil/rounding_mode.h"16#include "src/__support/common.h"17#include "src/__support/macros/config.h"18#include "src/__support/macros/optimization.h"19#include "src/__support/math/expxf16_utils.h"20 21namespace LIBC_NAMESPACE_DECL {22 23#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS24static constexpr fputil::ExceptValues<float16, 17> SINHF16_EXCEPTS_POS = {{25    // x = 0x1.714p-5, sinhf16(x) = 0x1.714p-5 (RZ)26    {0x29c5U, 0x29c5U, 1U, 0U, 1U},27    // x = 0x1.25p-4, sinhf16(x) = 0x1.25p-4 (RZ)28    {0x2c94U, 0x2c94U, 1U, 0U, 1U},29    // x = 0x1.f5p-4, sinhf16(x) = 0x1.f64p-4 (RZ)30    {0x2fd4U, 0x2fd9U, 1U, 0U, 0U},31    // x = 0x1.b1cp-3, sinhf16(x) = 0x1.b4cp-3 (RZ)32    {0x32c7U, 0x32d3U, 1U, 0U, 1U},33    // x = 0x1.6e8p-2, sinhf16(x) = 0x1.764p-2 (RZ)34    {0x35baU, 0x35d9U, 1U, 0U, 1U},35    // x = 0x1.6b4p-1, sinhf16(x) = 0x1.8a4p-1 (RZ)36    {0x39adU, 0x3a29U, 1U, 0U, 1U},37    // x = 0x1.a58p-1, sinhf16(x) = 0x1.d68p-1 (RZ)38    {0x3a96U, 0x3b5aU, 1U, 0U, 1U},39    // x = 0x1.574p+0, sinhf16(x) = 0x1.c78p+0 (RZ)40    {0x3d5dU, 0x3f1eU, 1U, 0U, 1U},41    // x = 0x1.648p+1, sinhf16(x) = 0x1.024p+3 (RZ)42    {0x4192U, 0x4809U, 1U, 0U, 0U},43    // x = 0x1.cdcp+1, sinhf16(x) = 0x1.26cp+4 (RZ)44    {0x4337U, 0x4c9bU, 1U, 0U, 0U},45    // x = 0x1.d0cp+1, sinhf16(x) = 0x1.2d8p+4 (RZ)46    {0x4343U, 0x4cb6U, 1U, 0U, 1U},47    // x = 0x1.018p+2, sinhf16(x) = 0x1.bfp+4 (RZ)48    {0x4406U, 0x4efcU, 1U, 0U, 0U},49    // x = 0x1.2fcp+2, sinhf16(x) = 0x1.cc4p+5 (RZ)50    {0x44bfU, 0x5331U, 1U, 0U, 1U},51    // x = 0x1.4ecp+2, sinhf16(x) = 0x1.75cp+6 (RZ)52    {0x453bU, 0x55d7U, 1U, 0U, 0U},53    // x = 0x1.8a4p+2, sinhf16(x) = 0x1.d94p+7 (RZ)54    {0x4629U, 0x5b65U, 1U, 0U, 1U},55    // x = 0x1.5fp+3, sinhf16(x) = 0x1.c54p+14 (RZ)56    {0x497cU, 0x7715U, 1U, 0U, 1U},57    // x = 0x1.3c8p+1, sinhf16(x) = 0x1.78ap+2 (RZ)58    {0x40f2U, 0x45e2U, 1U, 0U, 1U},59}};60 61static constexpr fputil::ExceptValues<float16, 13> SINHF16_EXCEPTS_NEG = {{62    // x = -0x1.714p-5, sinhf16(x) = -0x1.714p-5 (RZ)63    {0xa9c5U, 0xa9c5U, 0U, 1U, 1U},64    // x = -0x1.25p-4, sinhf16(x) = -0x1.25p-4 (RZ)65    {0xac94U, 0xac94U, 0U, 1U, 1U},66    // x = -0x1.f5p-4, sinhf16(x) = -0x1.f64p-4 (RZ)67    {0xafd4U, 0xafd9U, 0U, 1U, 0U},68    // x = -0x1.6e8p-2, sinhf16(x) = -0x1.764p-2 (RZ)69    {0xb5baU, 0xb5d9U, 0U, 1U, 1U},70    // x = -0x1.a58p-1, sinhf16(x) = -0x1.d68p-1 (RZ)71    {0xba96U, 0xbb5aU, 0U, 1U, 1U},72    // x = -0x1.cdcp+1, sinhf16(x) = -0x1.26cp+4 (RZ)73    {0xc337U, 0xcc9bU, 0U, 1U, 0U},74    // x = -0x1.d0cp+1, sinhf16(x) = -0x1.2d8p+4 (RZ)75    {0xc343U, 0xccb6U, 0U, 1U, 1U},76    // x = -0x1.018p+2, sinhf16(x) = -0x1.bfp+4 (RZ)77    {0xc406U, 0xcefcU, 0U, 1U, 0U},78    // x = -0x1.2fcp+2, sinhf16(x) = -0x1.cc4p+5 (RZ)79    {0xc4bfU, 0xd331U, 0U, 1U, 1U},80    // x = -0x1.4ecp+2, sinhf16(x) = -0x1.75cp+6 (RZ)81    {0xc53bU, 0xd5d7U, 0U, 1U, 0U},82    // x = -0x1.8a4p+2, sinhf16(x) = -0x1.d94p+7 (RZ)83    {0xc629U, 0xdb65U, 0U, 1U, 1U},84    // x = -0x1.5fp+3, sinhf16(x) = -0x1.c54p+14 (RZ)85    {0xc97cU, 0xf715U, 0U, 1U, 1U},86    // x = -0x1.3c8p+1, sinhf16(x) = -0x1.78ap+2 (RZ)87    {0xc0f2U, 0xc5e2U, 0U, 1U, 1U},88}};89#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS90 91LLVM_LIBC_FUNCTION(float16, sinhf16, (float16 x)) {92  using namespace math::expxf16_internal;93  using FPBits = fputil::FPBits<float16>;94  FPBits x_bits(x);95 96  uint16_t x_u = x_bits.uintval();97  uint16_t x_abs = x_u & 0x7fffU;98 99  // When |x| = 0, or -2^(-14) <= x <= -2^(-9), or |x| >= asinh(2^16), or x is100  // NaN.101  if (LIBC_UNLIKELY(x_abs == 0U || (x_u >= 0x8400U && x_u <= 0xa400U) ||102                    x_abs >= 0x49e5U)) {103    // sinh(NaN) = NaN104    if (x_bits.is_nan()) {105      if (x_bits.is_signaling_nan()) {106        fputil::raise_except_if_required(FE_INVALID);107        return FPBits::quiet_nan().get_val();108      }109 110      return x;111    }112 113    // sinh(+/-0) = sinh(+/-0)114    if (x_abs == 0U)115      return FPBits::zero(x_bits.sign()).get_val();116 117    // When |x| >= asinh(2^16).118    if (x_abs >= 0x49e5U) {119      // sinh(+/-inf) = +/-inf120      if (x_bits.is_inf())121        return FPBits::inf(x_bits.sign()).get_val();122 123      int rounding_mode = fputil::quick_get_round();124      if (rounding_mode == FE_TONEAREST ||125          (x_bits.is_pos() && rounding_mode == FE_UPWARD) ||126          (x_bits.is_neg() && rounding_mode == FE_DOWNWARD)) {127        fputil::set_errno_if_required(ERANGE);128        fputil::raise_except_if_required(FE_OVERFLOW | FE_INEXACT);129        return FPBits::inf(x_bits.sign()).get_val();130      }131      return FPBits::max_normal(x_bits.sign()).get_val();132    }133 134    // When -2^(-14) <= x <= -2^(-9).135    if (fputil::fenv_is_round_down())136      return FPBits(static_cast<uint16_t>(x_u + 1)).get_val();137    return FPBits(static_cast<uint16_t>(x_u)).get_val();138  }139 140#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS141  if (x_bits.is_pos()) {142    if (auto r = SINHF16_EXCEPTS_POS.lookup(x_u); LIBC_UNLIKELY(r.has_value()))143      return r.value();144  } else {145    if (auto r = SINHF16_EXCEPTS_NEG.lookup(x_u); LIBC_UNLIKELY(r.has_value()))146      return r.value();147  }148#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS149 150  return eval_sinh_or_cosh</*IsSinh=*/true>(x);151}152 153} // namespace LIBC_NAMESPACE_DECL154