1/* 2 * Copyright (c) 2014 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a copy 5 * of this software and associated documentation files (the "Software"), to deal 6 * in the Software without restriction, including without limitation the rights 7 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 8 * copies of the Software, and to permit persons to whom the Software is 9 * furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 17 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 19 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 20 * THE SOFTWARE. 21 */ 22 23#include <clc/clc.h> 24#include <clc/clcmacro.h> 25#include <clc/integer/clc_add_sat.h> 26#include <clc/math/clc_subnormal_config.h> 27#include <clc/math/math.h> 28#include <clc/relational/clc_isinf.h> 29#include <clc/relational/clc_isnan.h> 30#include <clc/shared/clc_clamp.h> 31 32_CLC_DEF _CLC_OVERLOAD float __clc_ldexp(float x, int n) { 33 34 if (!__clc_fp32_subnormals_supported()) { 35 36 // This treats subnormals as zeros 37 int i = as_int(x); 38 int e = (i >> 23) & 0xff; 39 int m = i & 0x007fffff; 40 int s = i & 0x80000000; 41 int v = __clc_add_sat(e, n); 42 v = __clc_clamp(v, 0, 0xff); 43 int mr = e == 0 | v == 0 | v == 0xff ? 0 : m; 44 int c = e == 0xff; 45 mr = c ? m : mr; 46 int er = c ? e : v; 47 er = e ? er : e; 48 return as_float(s | (er << 23) | mr); 49 } 50 51 /* supports denormal values */ 52 const int multiplier = 24; 53 float val_f; 54 uint val_ui; 55 uint sign; 56 int exponent; 57 val_ui = as_uint(x); 58 sign = val_ui & 0x80000000; 59 val_ui = val_ui & 0x7fffffff; /* remove the sign bit */ 60 int val_x = val_ui; 61 62 exponent = val_ui >> 23; /* get the exponent */ 63 int dexp = exponent; 64 65 /* denormal support */ 66 int fbh = 67 127 - (as_uint((float)(as_float(val_ui | 0x3f800000) - 1.0f)) >> 23); 68 int dexponent = 25 - fbh; 69 uint dval_ui = (((val_ui << fbh) & 0x007fffff) | (dexponent << 23)); 70 int ex = dexponent + n - multiplier; 71 dexponent = ex; 72 uint val = sign | (ex << 23) | (dval_ui & 0x007fffff); 73 int ex1 = dexponent + multiplier; 74 ex1 = -ex1 + 25; 75 dval_ui = (((dval_ui & 0x007fffff) | 0x800000) >> ex1); 76 dval_ui = dexponent > 0 ? val : dval_ui; 77 dval_ui = dexponent > 254 ? 0x7f800000 : dval_ui; /*overflow*/ 78 dval_ui = dexponent < -multiplier ? 0 : dval_ui; /*underflow*/ 79 dval_ui = dval_ui | sign; 80 val_f = as_float(dval_ui); 81 82 exponent += n; 83 84 val = sign | (exponent << 23) | (val_ui & 0x007fffff); 85 ex1 = exponent + multiplier; 86 ex1 = -ex1 + 25; 87 val_ui = (((val_ui & 0x007fffff) | 0x800000) >> ex1); 88 val_ui = exponent > 0 ? val : val_ui; 89 val_ui = exponent > 254 ? 0x7f800000 : val_ui; /*overflow*/ 90 val_ui = exponent < -multiplier ? 0 : val_ui; /*underflow*/ 91 val_ui = val_ui | sign; 92 93 val_ui = dexp == 0 ? dval_ui : val_ui; 94 val_f = as_float(val_ui); 95 96 val_f = __clc_isnan(x) | __clc_isinf(x) | val_x == 0 ? x : val_f; 97 return val_f; 98} 99 100#ifdef cl_khr_fp64 101 102#pragma OPENCL EXTENSION cl_khr_fp64 : enable 103 104_CLC_DEF _CLC_OVERLOAD double __clc_ldexp(double x, int n) { 105 long l = as_ulong(x); 106 int e = (l >> 52) & 0x7ff; 107 long s = l & 0x8000000000000000; 108 109 ulong ux = as_ulong(x * 0x1.0p+53); 110 int de = ((int)(ux >> 52) & 0x7ff) - 53; 111 int c = e == 0; 112 e = c ? de : e; 113 114 ux = c ? ux : l; 115 116 int v = e + n; 117 v = __clc_clamp(v, -0x7ff, 0x7ff); 118 119 ux &= ~EXPBITS_DP64; 120 121 double mr = as_double(ux | ((ulong)(v + 53) << 52)); 122 mr = mr * 0x1.0p-53; 123 124 mr = v > 0 ? as_double(ux | ((ulong)v << 52)) : mr; 125 126 mr = v == 0x7ff ? as_double(s | PINFBITPATT_DP64) : mr; 127 mr = v < -53 ? as_double(s) : mr; 128 129 mr = ((n == 0) | __clc_isinf(x) | (x == 0)) ? x : mr; 130 return mr; 131} 132 133#endif 134 135#ifdef cl_khr_fp16 136 137#pragma OPENCL EXTENSION cl_khr_fp16 : enable 138 139_CLC_OVERLOAD _CLC_DEF half __clc_ldexp(half x, int n) { 140 return (half)__clc_ldexp((float)x, n); 141} 142 143_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __clc_ldexp, half, int); 144 145#endif 146