//===-- Half-precision e^x function ---------------------------------------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// #include "src/math/expf16.h" #include "expxf16.h" #include "hdr/errno_macros.h" #include "hdr/fenv_macros.h" #include "src/__support/FPUtil/FEnvImpl.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/FPUtil/PolyEval.h" #include "src/__support/FPUtil/cast.h" #include "src/__support/FPUtil/except_value_utils.h" #include "src/__support/FPUtil/rounding_mode.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" #include "src/__support/macros/optimization.h" namespace LIBC_NAMESPACE_DECL { static constexpr fputil::ExceptValues EXPF16_EXCEPTS_LO = {{ // (input, RZ output, RU offset, RD offset, RN offset) // x = 0x1.de4p-8, expf16(x) = 0x1.01cp+0 (RZ) {0x1f79U, 0x3c07U, 1U, 0U, 0U}, // x = 0x1.73cp-6, expf16(x) = 0x1.05cp+0 (RZ) {0x25cfU, 0x3c17U, 1U, 0U, 0U}, }}; static constexpr fputil::ExceptValues EXPF16_EXCEPTS_HI = {{ // (input, RZ output, RU offset, RD offset, RN offset) // x = 0x1.c34p+0, expf16(x) = 0x1.74cp+2 (RZ) {0x3f0dU, 0x45d3U, 1U, 0U, 1U}, // x = -0x1.488p-5, expf16(x) = 0x1.ebcp-1 (RZ) {0xa922U, 0x3bafU, 1U, 0U, 0U}, // x = -0x1.55p-5, expf16(x) = 0x1.ebp-1 (RZ) {0xa954U, 0x3bacU, 1U, 0U, 0U}, }}; LLVM_LIBC_FUNCTION(float16, expf16, (float16 x)) { using FPBits = fputil::FPBits; FPBits x_bits(x); uint16_t x_u = x_bits.uintval(); uint16_t x_abs = x_u & 0x7fffU; // When 0 < |x| <= 2^(-5), or |x| >= 12, or x is NaN. if (LIBC_UNLIKELY(x_abs <= 0x2800U || x_abs >= 0x4a00U)) { // exp(NaN) = NaN if (x_bits.is_nan()) { if (x_bits.is_signaling_nan()) { fputil::raise_except_if_required(FE_INVALID); return FPBits::quiet_nan().get_val(); } return x; } // When x >= 12. if (x_bits.is_pos() && x_abs >= 0x4a00U) { // exp(+inf) = +inf if (x_bits.is_inf()) return FPBits::inf().get_val(); switch (fputil::quick_get_round()) { case FE_TONEAREST: case FE_UPWARD: fputil::set_errno_if_required(ERANGE); fputil::raise_except_if_required(FE_OVERFLOW); return FPBits::inf().get_val(); default: return FPBits::max_normal().get_val(); } } // When x <= -18. if (x_u >= 0xcc80U) { // exp(-inf) = +0 if (x_bits.is_inf()) return FPBits::zero().get_val(); fputil::set_errno_if_required(ERANGE); fputil::raise_except_if_required(FE_UNDERFLOW | FE_INEXACT); switch (fputil::quick_get_round()) { case FE_UPWARD: return FPBits::min_subnormal().get_val(); default: return FPBits::zero().get_val(); } } // When 0 < |x| <= 2^(-5). if (x_abs <= 0x2800U && !x_bits.is_zero()) { if (auto r = EXPF16_EXCEPTS_LO.lookup(x_u); LIBC_UNLIKELY(r.has_value())) return r.value(); float xf = x; // Degree-3 minimax polynomial generated by Sollya with the following // commands: // > display = hexadecimal; // > P = fpminimax(expm1(x)/x, 2, [|SG...|], [-2^-5, 2^-5]); // > 1 + x * P; return fputil::cast( fputil::polyeval(xf, 0x1p+0f, 0x1p+0f, 0x1.0004p-1f, 0x1.555778p-3f)); } } if (auto r = EXPF16_EXCEPTS_HI.lookup(x_u); LIBC_UNLIKELY(r.has_value())) return r.value(); // exp(x) = exp(hi + mid) * exp(lo) auto [exp_hi_mid, exp_lo] = exp_range_reduction(x); return fputil::cast(exp_hi_mid * exp_lo); } } // namespace LIBC_NAMESPACE_DECL