77 lines · plain
1//===----------------------------------------------------------------------===//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#if __CLC_FPSIZE == 3210 11_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp(__CLC_GENTYPE x) {12 // Reduce x13 const __CLC_GENTYPE ln2HI = 0x1.62e300p-1f;14 const __CLC_GENTYPE ln2LO = 0x1.2fefa2p-17f;15 const __CLC_GENTYPE invln2 = 0x1.715476p+0f;16 17 __CLC_GENTYPE fhalF = x < 0.0f ? -0.5f : 0.5f;18 __CLC_INTN p = __CLC_CONVERT_INTN(__clc_mad(x, invln2, fhalF));19 __CLC_GENTYPE fp = __CLC_CONVERT_GENTYPE(p);20 __CLC_GENTYPE hi = __clc_mad(fp, -ln2HI, x); // t*ln2HI is exact here21 __CLC_GENTYPE lo = -fp * ln2LO;22 23 // Evaluate poly24 __CLC_GENTYPE t = hi + lo;25 __CLC_GENTYPE tt = t * t;26 __CLC_GENTYPE v = __clc_mad(27 tt,28 -__clc_mad(29 tt,30 __clc_mad(tt,31 __clc_mad(tt,32 __clc_mad(tt, 0x1.637698p-25f, -0x1.bbd41cp-20f),33 0x1.1566aap-14f),34 -0x1.6c16c2p-9f),35 0x1.555556p-3f),36 t);37 38 __CLC_GENTYPE y = 1.0f - (((-lo) - MATH_DIVIDE(t * v, 2.0f - v)) - hi);39 40 // Scale by 2^p41 __CLC_GENTYPE r = __CLC_AS_GENTYPE(__CLC_AS_INTN(y) + (p << 23));42 43 // ln(largest_normal) = 88.7228390520683530536644 const __CLC_GENTYPE ulim = 0x1.62e430p+6f;45 // ln(smallest_normal) = -87.3365447505531089865746 const __CLC_GENTYPE llim = -0x1.5d589ep+6f;47 48 r = x < llim ? 0.0f : r;49 r = x < ulim ? r : __CLC_AS_GENTYPE((__CLC_UINTN)0x7f800000);50 return __clc_isnan(x) ? x : r;51}52 53#elif __CLC_FPSIZE == 6454 55_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp(__CLC_GENTYPE x) {56 57 const __CLC_GENTYPE X_MIN = -0x1.74910d52d3051p+9; // -1075*ln(2)58 const __CLC_GENTYPE X_MAX = 0x1.62e42fefa39efp+9; // 1024*ln(2)59 const __CLC_GENTYPE R_64_BY_LOG2 = 0x1.71547652b82fep+6; // 64/ln(2)60 const __CLC_GENTYPE R_LOG2_BY_64_LD = 0x1.62e42fefa0000p-7; // head ln(2)/6461 const __CLC_GENTYPE R_LOG2_BY_64_TL = 0x1.cf79abc9e3b39p-46; // tail ln(2)/6462 63 __CLC_INTN n = __CLC_CONVERT_INTN(x * R_64_BY_LOG2);64 __CLC_GENTYPE r =65 __clc_fma(-R_LOG2_BY_64_TL, __CLC_CONVERT_GENTYPE(n),66 __clc_fma(-R_LOG2_BY_64_LD, __CLC_CONVERT_GENTYPE(n), x));67 return __clc_exp_helper(x, X_MIN, X_MAX, r, n);68}69 70#elif __CLC_FPSIZE == 1671 72_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp(__CLC_GENTYPE x) {73 return __CLC_CONVERT_GENTYPE(__clc_exp(__CLC_CONVERT_FLOATN(x)));74}75 76#endif77