| 1 | /* Single-precision e^x function. |
| 2 | Copyright (C) 2017-2022 Free Software Foundation, Inc. |
| 3 | This file is part of the GNU C Library. |
| 4 | |
| 5 | The GNU C Library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with the GNU C Library; if not, see |
| 17 | <https://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #ifdef __expf |
| 20 | # undef libm_hidden_proto |
| 21 | # define libm_hidden_proto(ignored) |
| 22 | #endif |
| 23 | |
| 24 | #include <math.h> |
| 25 | #include <math-narrow-eval.h> |
| 26 | #include <stdint.h> |
| 27 | #include <libm-alias-finite.h> |
| 28 | #include <libm-alias-float.h> |
| 29 | #include "math_config.h" |
| 30 | |
| 31 | /* |
| 32 | EXP2F_TABLE_BITS = 5 |
| 33 | EXP2F_POLY_ORDER = 3 |
| 34 | |
| 35 | ULP error: 0.502 (nearest rounding.) |
| 36 | Relative error: 1.69 * 2^-34 in [-ln2/64, ln2/64] (before rounding.) |
| 37 | Wrong count: 170635 (all nearest rounding wrong results with fma.) |
| 38 | Non-nearest ULP error: 1 (rounded ULP error) |
| 39 | */ |
| 40 | |
| 41 | #define N (1 << EXP2F_TABLE_BITS) |
| 42 | #define InvLn2N __exp2f_data.invln2_scaled |
| 43 | #define T __exp2f_data.tab |
| 44 | #define C __exp2f_data.poly_scaled |
| 45 | |
| 46 | static inline uint32_t |
| 47 | top12 (float x) |
| 48 | { |
| 49 | return asuint (x) >> 20; |
| 50 | } |
| 51 | |
| 52 | float |
| 53 | __expf (float x) |
| 54 | { |
| 55 | uint32_t abstop; |
| 56 | uint64_t ki, t; |
| 57 | /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */ |
| 58 | double_t kd, xd, z, r, r2, y, s; |
| 59 | |
| 60 | xd = (double_t) x; |
| 61 | abstop = top12 (x) & 0x7ff; |
| 62 | if (__glibc_unlikely (abstop >= top12 (88.0f))) |
| 63 | { |
| 64 | /* |x| >= 88 or x is nan. */ |
| 65 | if (asuint (x) == asuint (-INFINITY)) |
| 66 | return 0.0f; |
| 67 | if (abstop >= top12 (INFINITY)) |
| 68 | return x + x; |
| 69 | if (x > 0x1.62e42ep6f) /* x > log(0x1p128) ~= 88.72 */ |
| 70 | return __math_oflowf (0); |
| 71 | if (x < -0x1.9fe368p6f) /* x < log(0x1p-150) ~= -103.97 */ |
| 72 | return __math_uflowf (0); |
| 73 | #if WANT_ERRNO_UFLOW |
| 74 | if (x < -0x1.9d1d9ep6f) /* x < log(0x1p-149) ~= -103.28 */ |
| 75 | return __math_may_uflowf (0); |
| 76 | #endif |
| 77 | } |
| 78 | |
| 79 | /* x*N/Ln2 = k + r with r in [-1/2, 1/2] and int k. */ |
| 80 | z = InvLn2N * xd; |
| 81 | |
| 82 | /* Round and convert z to int, the result is in [-150*N, 128*N] and |
| 83 | ideally ties-to-even rule is used, otherwise the magnitude of r |
| 84 | can be bigger which gives larger approximation error. */ |
| 85 | #if TOINT_INTRINSICS |
| 86 | kd = roundtoint (z); |
| 87 | ki = converttoint (z); |
| 88 | #else |
| 89 | # define SHIFT __exp2f_data.shift |
| 90 | kd = math_narrow_eval ((double) (z + SHIFT)); /* Needs to be double. */ |
| 91 | ki = asuint64 (kd); |
| 92 | kd -= SHIFT; |
| 93 | #endif |
| 94 | r = z - kd; |
| 95 | |
| 96 | /* exp(x) = 2^(k/N) * 2^(r/N) ~= s * (C0*r^3 + C1*r^2 + C2*r + 1) */ |
| 97 | t = T[ki % N]; |
| 98 | t += ki << (52 - EXP2F_TABLE_BITS); |
| 99 | s = asdouble (t); |
| 100 | z = C[0] * r + C[1]; |
| 101 | r2 = r * r; |
| 102 | y = C[2] * r + 1; |
| 103 | y = z * r2 + y; |
| 104 | y = y * s; |
| 105 | return (float) y; |
| 106 | } |
| 107 | |
| 108 | #ifndef __expf |
| 109 | hidden_def (__expf) |
| 110 | strong_alias (__expf, __ieee754_expf) |
| 111 | libm_alias_finite (__ieee754_expf, __expf) |
| 112 | versioned_symbol (libm, __expf, expf, GLIBC_2_27); |
| 113 | libm_alias_float_other (__exp, exp) |
| 114 | #endif |
| 115 | |