toilscript 0.0.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +201 -0
- package/NOTICE +94 -0
- package/README.md +114 -0
- package/bin/asc.js +35 -0
- package/bin/asinit.js +468 -0
- package/dist/asc.d.ts +4 -0
- package/dist/toilscript.d.ts +4 -0
- package/dist/transform.cjs +1 -0
- package/dist/transform.d.ts +1 -0
- package/dist/transform.js +1 -0
- package/lib/binaryen.d.ts +2 -0
- package/lib/binaryen.js +2 -0
- package/package.json +114 -0
- package/std/README.md +6 -0
- package/std/assembly/array.ts +550 -0
- package/std/assembly/arraybuffer.ts +77 -0
- package/std/assembly/atomics.ts +127 -0
- package/std/assembly/bindings/asyncify.ts +16 -0
- package/std/assembly/bindings/dom.ts +291 -0
- package/std/assembly/bindings/node.ts +6 -0
- package/std/assembly/bitflags.ts +53 -0
- package/std/assembly/builtins.ts +2650 -0
- package/std/assembly/byteslice.ts +177 -0
- package/std/assembly/compat.ts +2 -0
- package/std/assembly/console.ts +42 -0
- package/std/assembly/crypto.ts +9 -0
- package/std/assembly/dataview.ts +181 -0
- package/std/assembly/date.ts +375 -0
- package/std/assembly/diagnostics.ts +11 -0
- package/std/assembly/encoding.ts +151 -0
- package/std/assembly/endian.ts +45 -0
- package/std/assembly/error.ts +44 -0
- package/std/assembly/fixedarray.ts +173 -0
- package/std/assembly/fixedmap.ts +326 -0
- package/std/assembly/fixedset.ts +275 -0
- package/std/assembly/function.ts +42 -0
- package/std/assembly/index.d.ts +2891 -0
- package/std/assembly/iterator.ts +35 -0
- package/std/assembly/map.ts +269 -0
- package/std/assembly/math.ts +3289 -0
- package/std/assembly/memory.ts +123 -0
- package/std/assembly/number.ts +388 -0
- package/std/assembly/object.ts +36 -0
- package/std/assembly/performance.ts +9 -0
- package/std/assembly/pointer.ts +80 -0
- package/std/assembly/polyfills.ts +27 -0
- package/std/assembly/process.ts +50 -0
- package/std/assembly/reference.ts +48 -0
- package/std/assembly/regexp.ts +12 -0
- package/std/assembly/rt/README.md +83 -0
- package/std/assembly/rt/common.ts +81 -0
- package/std/assembly/rt/index-incremental.ts +2 -0
- package/std/assembly/rt/index-memory.ts +1 -0
- package/std/assembly/rt/index-minimal.ts +2 -0
- package/std/assembly/rt/index-stub.ts +1 -0
- package/std/assembly/rt/index.d.ts +37 -0
- package/std/assembly/rt/itcms.ts +419 -0
- package/std/assembly/rt/memory-runtime.ts +94 -0
- package/std/assembly/rt/rtrace.ts +15 -0
- package/std/assembly/rt/stub.ts +133 -0
- package/std/assembly/rt/tcms.ts +254 -0
- package/std/assembly/rt/tlsf.ts +592 -0
- package/std/assembly/rt.ts +90 -0
- package/std/assembly/set.ts +225 -0
- package/std/assembly/shared/feature.ts +68 -0
- package/std/assembly/shared/runtime.ts +13 -0
- package/std/assembly/shared/target.ts +11 -0
- package/std/assembly/shared/tsconfig.json +11 -0
- package/std/assembly/shared/typeinfo.ts +72 -0
- package/std/assembly/staticarray.ts +423 -0
- package/std/assembly/string.ts +850 -0
- package/std/assembly/symbol.ts +114 -0
- package/std/assembly/table.ts +16 -0
- package/std/assembly/tsconfig.json +6 -0
- package/std/assembly/typedarray.ts +1954 -0
- package/std/assembly/uri.ts +17 -0
- package/std/assembly/util/bytes.ts +107 -0
- package/std/assembly/util/casemap.ts +497 -0
- package/std/assembly/util/error.ts +58 -0
- package/std/assembly/util/hash.ts +117 -0
- package/std/assembly/util/math.ts +1922 -0
- package/std/assembly/util/memory.ts +290 -0
- package/std/assembly/util/number.ts +873 -0
- package/std/assembly/util/sort.ts +313 -0
- package/std/assembly/util/string.ts +1202 -0
- package/std/assembly/util/uri.ts +275 -0
- package/std/assembly/vector.ts +4 -0
- package/std/assembly.json +16 -0
- package/std/portable/index.d.ts +461 -0
- package/std/portable/index.js +416 -0
- package/std/portable.json +11 -0
- package/std/types/assembly/index.d.ts +1 -0
- package/std/types/assembly/package.json +3 -0
- package/std/types/portable/index.d.ts +1 -0
- package/std/types/portable/package.json +3 -0
- package/tsconfig-base.json +13 -0
- package/util/README.md +23 -0
- package/util/browser/fs.js +1 -0
- package/util/browser/module.js +5 -0
- package/util/browser/path.js +520 -0
- package/util/browser/process.js +59 -0
- package/util/browser/url.js +23 -0
- package/util/cpu.d.ts +9 -0
- package/util/cpu.js +42 -0
- package/util/find.d.ts +6 -0
- package/util/find.js +20 -0
- package/util/node.d.ts +21 -0
- package/util/node.js +34 -0
- package/util/options.d.ts +70 -0
- package/util/options.js +262 -0
- package/util/terminal.d.ts +52 -0
- package/util/terminal.js +35 -0
- package/util/text.d.ts +26 -0
- package/util/text.js +114 -0
- package/util/tsconfig.json +9 -0
- package/util/web.d.ts +11 -0
- package/util/web.js +33 -0
|
@@ -0,0 +1,3289 @@
|
|
|
1
|
+
import { Math as JSMath } from "./bindings/dom";
|
|
2
|
+
export { JSMath };
|
|
3
|
+
|
|
4
|
+
import {
|
|
5
|
+
pow_lut, exp_lut, exp2_lut, log_lut, log2_lut,
|
|
6
|
+
powf_lut, expf_lut, exp2f_lut, logf_lut, log2f_lut
|
|
7
|
+
} from "./util/math";
|
|
8
|
+
|
|
9
|
+
import {
|
|
10
|
+
abs as builtin_abs,
|
|
11
|
+
ceil as builtin_ceil,
|
|
12
|
+
clz as builtin_clz,
|
|
13
|
+
copysign as builtin_copysign,
|
|
14
|
+
floor as builtin_floor,
|
|
15
|
+
max as builtin_max,
|
|
16
|
+
min as builtin_min,
|
|
17
|
+
sqrt as builtin_sqrt,
|
|
18
|
+
trunc as builtin_trunc
|
|
19
|
+
} from "./builtins";
|
|
20
|
+
|
|
21
|
+
// SUN COPYRIGHT NOTICE
|
|
22
|
+
//
|
|
23
|
+
// Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
|
|
24
|
+
// Developed at SunPro, a Sun Microsystems, Inc. business.
|
|
25
|
+
// Permission to use, copy, modify, and distribute this software
|
|
26
|
+
// is freely granted, provided that this notice is preserved.
|
|
27
|
+
//
|
|
28
|
+
// Applies to all functions marked with a comment referring here.
|
|
29
|
+
|
|
30
|
+
/** @internal */
|
|
31
|
+
// @ts-ignore: decorator
|
|
32
|
+
@lazy let rempio2_y0: f64, rempio2_y1: f64, res128_hi: u64;
|
|
33
|
+
|
|
34
|
+
/** @internal */
|
|
35
|
+
// @ts-ignore: decorator
|
|
36
|
+
@lazy @inline const PIO2_TABLE = memory.data<u64>([
|
|
37
|
+
0x00000000A2F9836E, 0x4E441529FC2757D1, 0xF534DDC0DB629599, 0x3C439041FE5163AB,
|
|
38
|
+
0xDEBBC561B7246E3A, 0x424DD2E006492EEA, 0x09D1921CFE1DEB1C, 0xB129A73EE88235F5,
|
|
39
|
+
0x2EBB4484E99C7026, 0xB45F7E413991D639, 0x835339F49C845F8B, 0xBDF9283B1FF897FF,
|
|
40
|
+
0xDE05980FEF2F118B, 0x5A0A6D1F6D367ECF, 0x27CB09B74F463F66, 0x9E5FEA2D7527BAC7,
|
|
41
|
+
0xEBE5F17B3D0739F7, 0x8A5292EA6BFB5FB1, 0x1F8D5D0856033046, 0xFC7B6BABF0CFBC20,
|
|
42
|
+
0x9AF4361DA9E39161, 0x5EE61B086599855F, 0x14A068408DFFD880, 0x4D73273106061557
|
|
43
|
+
]);
|
|
44
|
+
|
|
45
|
+
/** @internal */
|
|
46
|
+
function R(z: f64): f64 { // Rational approximation of (asin(x)-x)/x^3
|
|
47
|
+
const // see: musl/src/math/asin.c and SUN COPYRIGHT NOTICE above
|
|
48
|
+
pS0 = reinterpret<f64>(0x3FC5555555555555), // 1.66666666666666657415e-01
|
|
49
|
+
pS1 = reinterpret<f64>(0xBFD4D61203EB6F7D), // -3.25565818622400915405e-01
|
|
50
|
+
pS2 = reinterpret<f64>(0x3FC9C1550E884455), // 2.01212532134862925881e-01
|
|
51
|
+
pS3 = reinterpret<f64>(0xBFA48228B5688F3B), // -4.00555345006794114027e-02
|
|
52
|
+
pS4 = reinterpret<f64>(0x3F49EFE07501B288), // 7.91534994289814532176e-04
|
|
53
|
+
pS5 = reinterpret<f64>(0x3F023DE10DFDF709), // 3.47933107596021167570e-05
|
|
54
|
+
qS1 = reinterpret<f64>(0xC0033A271C8A2D4B), // -2.40339491173441421878e+00
|
|
55
|
+
qS2 = reinterpret<f64>(0x40002AE59C598AC8), // 2.02094576023350569471e+00
|
|
56
|
+
qS3 = reinterpret<f64>(0xBFE6066C1B8D0159), // -6.88283971605453293030e-01
|
|
57
|
+
qS4 = reinterpret<f64>(0x3FB3B8C5B12E9282); // 7.70381505559019352791e-02
|
|
58
|
+
|
|
59
|
+
let p = z * (pS0 + z * (pS1 + z * (pS2 + z * (pS3 + z * (pS4 + z * pS5)))));
|
|
60
|
+
let q = 1.0 + z * (qS1 + z * (qS2 + z * (qS3 + z * qS4)));
|
|
61
|
+
return p / q;
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
/** @internal */
|
|
65
|
+
// @ts-ignore: decorator
|
|
66
|
+
@inline
|
|
67
|
+
function expo2(x: f64, sign: f64): f64 { // exp(x)/2 for x >= log(DBL_MAX)
|
|
68
|
+
const // see: musl/src/math/__expo2.c
|
|
69
|
+
k = <u32>2043,
|
|
70
|
+
kln2 = reinterpret<f64>(0x40962066151ADD8B); // 0x1.62066151add8bp+10
|
|
71
|
+
let scale = reinterpret<f64>(<u64>((<u32>0x3FF + k / 2) << 20) << 32);
|
|
72
|
+
// in directed rounding correct sign before rounding or overflow is important
|
|
73
|
+
return NativeMath.exp(x - kln2) * (sign * scale) * scale;
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
/** @internal */
|
|
77
|
+
/* Helper function to eventually get bits of π/2 * |x|
|
|
78
|
+
*
|
|
79
|
+
* y = π/4 * (frac << clz(frac) >> 11)
|
|
80
|
+
* return clz(frac)
|
|
81
|
+
*
|
|
82
|
+
* Right shift 11 bits to make upper half fit in `double`
|
|
83
|
+
*/
|
|
84
|
+
// @ts-ignore: decorator
|
|
85
|
+
@inline
|
|
86
|
+
function pio2_right(q0: u64, q1: u64): u64 { // see: jdh8/metallic/blob/master/src/math/double/rem_pio2.c
|
|
87
|
+
// Bits of π/4
|
|
88
|
+
const p0: u64 = 0xC4C6628B80DC1CD1;
|
|
89
|
+
const p1: u64 = 0xC90FDAA22168C234;
|
|
90
|
+
|
|
91
|
+
const Ox1p_64 = reinterpret<f64>(0x3BF0000000000000); // 0x1p-64
|
|
92
|
+
const Ox1p_75 = reinterpret<f64>(0x3B40000000000000); // 0x1p-75
|
|
93
|
+
|
|
94
|
+
let shift = clz(q1);
|
|
95
|
+
|
|
96
|
+
q1 = q1 << shift | q0 >> (64 - shift);
|
|
97
|
+
q0 <<= shift;
|
|
98
|
+
|
|
99
|
+
let lo = umuldi(p1, q1);
|
|
100
|
+
let hi = res128_hi;
|
|
101
|
+
|
|
102
|
+
let ahi = hi >> 11;
|
|
103
|
+
let alo = lo >> 11 | hi << 53;
|
|
104
|
+
let blo = <u64>(Ox1p_75 * <f64>p0 * <f64>q1 + Ox1p_75 * <f64>p1 * <f64>q0);
|
|
105
|
+
|
|
106
|
+
rempio2_y0 = <f64>(ahi + u64(lo < blo));
|
|
107
|
+
rempio2_y1 = Ox1p_64 * <f64>(alo + blo);
|
|
108
|
+
|
|
109
|
+
return shift;
|
|
110
|
+
}
|
|
111
|
+
|
|
112
|
+
/** @internal */
|
|
113
|
+
// @ts-ignore: decorator
|
|
114
|
+
@inline
|
|
115
|
+
function umuldi(u: u64, v: u64): u64 {
|
|
116
|
+
let u1: u64 , v1: u64, w0: u64, w1: u64, t: u64;
|
|
117
|
+
|
|
118
|
+
u1 = u & 0xFFFFFFFF;
|
|
119
|
+
v1 = v & 0xFFFFFFFF;
|
|
120
|
+
|
|
121
|
+
u >>= 32;
|
|
122
|
+
v >>= 32;
|
|
123
|
+
|
|
124
|
+
t = u1 * v1;
|
|
125
|
+
w0 = t & 0xFFFFFFFF;
|
|
126
|
+
t = u * v1 + (t >> 32);
|
|
127
|
+
w1 = t >> 32;
|
|
128
|
+
t = u1 * v + (t & 0xFFFFFFFF);
|
|
129
|
+
|
|
130
|
+
res128_hi = u * v + w1 + (t >> 32);
|
|
131
|
+
return (t << 32) + w0;
|
|
132
|
+
}
|
|
133
|
+
|
|
134
|
+
/** @internal */
|
|
135
|
+
function pio2_large_quot(x: f64, u: i64): i32 { // see: jdh8/metallic/blob/master/src/math/double/rem_pio2.c
|
|
136
|
+
let magnitude = u & 0x7FFFFFFFFFFFFFFF;
|
|
137
|
+
let offset = (magnitude >> 52) - 1045;
|
|
138
|
+
let shift = offset & 63;
|
|
139
|
+
let tblPtr = PIO2_TABLE + (<i32>(offset >> 6) << 3);
|
|
140
|
+
let s0: u64, s1: u64, s2: u64;
|
|
141
|
+
|
|
142
|
+
let b0 = load<u64>(tblPtr, 0 << 3);
|
|
143
|
+
let b1 = load<u64>(tblPtr, 1 << 3);
|
|
144
|
+
let b2 = load<u64>(tblPtr, 2 << 3);
|
|
145
|
+
|
|
146
|
+
// Get 192 bits of 0x1p-31 / π with `offset` bits skipped
|
|
147
|
+
if (shift) {
|
|
148
|
+
let rshift = 64 - shift;
|
|
149
|
+
let b3 = load<u64>(tblPtr, 3 << 3);
|
|
150
|
+
s0 = b1 >> rshift | b0 << shift;
|
|
151
|
+
s1 = b2 >> rshift | b1 << shift;
|
|
152
|
+
s2 = b3 >> rshift | b2 << shift;
|
|
153
|
+
} else {
|
|
154
|
+
s0 = b0;
|
|
155
|
+
s1 = b1;
|
|
156
|
+
s2 = b2;
|
|
157
|
+
}
|
|
158
|
+
|
|
159
|
+
let significand = (u & 0x000FFFFFFFFFFFFF) | 0x0010000000000000;
|
|
160
|
+
|
|
161
|
+
// First 128 bits of fractional part of x/(2π)
|
|
162
|
+
let blo = umuldi(s1, significand);
|
|
163
|
+
let bhi = res128_hi;
|
|
164
|
+
|
|
165
|
+
let ahi = s0 * significand;
|
|
166
|
+
let clo = (s2 >> 32) * (significand >> 32);
|
|
167
|
+
let plo = blo + clo;
|
|
168
|
+
let phi = ahi + bhi + u64(plo < clo);
|
|
169
|
+
|
|
170
|
+
// r: u128 = p << 2
|
|
171
|
+
let rlo = plo << 2;
|
|
172
|
+
let rhi = phi << 2 | plo >> 62;
|
|
173
|
+
|
|
174
|
+
// s: i128 = r >> 127
|
|
175
|
+
let slo = <i64>rhi >> 63;
|
|
176
|
+
let shi = slo >> 1;
|
|
177
|
+
let q = (<i64>phi >> 62) - slo;
|
|
178
|
+
|
|
179
|
+
let shifter = 0x3CB0000000000000 - (pio2_right(rlo ^ slo, rhi ^ shi) << 52);
|
|
180
|
+
let signbit = (u ^ rhi) & 0x8000000000000000;
|
|
181
|
+
let coeff = reinterpret<f64>(shifter | signbit);
|
|
182
|
+
|
|
183
|
+
rempio2_y0 *= coeff;
|
|
184
|
+
rempio2_y1 *= coeff;
|
|
185
|
+
|
|
186
|
+
return <i32>q;
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
/** @internal */
|
|
190
|
+
// @ts-ignore: decorator
|
|
191
|
+
@inline
|
|
192
|
+
function rempio2(x: f64, u: u64, sign: i32): i32 {
|
|
193
|
+
const
|
|
194
|
+
pio2_1 = reinterpret<f64>(0x3FF921FB54400000), // 1.57079632673412561417e+00
|
|
195
|
+
pio2_1t = reinterpret<f64>(0x3DD0B4611A626331), // 6.07710050650619224932e-11
|
|
196
|
+
pio2_2 = reinterpret<f64>(0x3DD0B4611A600000), // 6.07710050630396597660e-11
|
|
197
|
+
pio2_2t = reinterpret<f64>(0x3BA3198A2E037073), // 2.02226624879595063154e-21
|
|
198
|
+
pio2_3 = reinterpret<f64>(0x3BA3198A2E000000), // 2.02226624871116645580e-21
|
|
199
|
+
pio2_3t = reinterpret<f64>(0x397B839A252049C1), // 8.47842766036889956997e-32
|
|
200
|
+
invpio2 = reinterpret<f64>(0x3FE45F306DC9C883); // 0.63661977236758134308
|
|
201
|
+
|
|
202
|
+
let ix = <u32>(u >> 32) & 0x7FFFFFFF;
|
|
203
|
+
|
|
204
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
205
|
+
if (ix < 0x4002D97C) { // |x| < 3pi/4, special case with n=+-1
|
|
206
|
+
let q = 1, z: f64, y0: f64, y1: f64;
|
|
207
|
+
if (!sign) {
|
|
208
|
+
z = x - pio2_1;
|
|
209
|
+
if (ix != 0x3FF921FB) { // 33+53 bit pi is good enough
|
|
210
|
+
y0 = z - pio2_1t;
|
|
211
|
+
y1 = (z - y0) - pio2_1t;
|
|
212
|
+
} else { // near pi/2, use 33+33+53 bit pi
|
|
213
|
+
z -= pio2_2;
|
|
214
|
+
y0 = z - pio2_2t;
|
|
215
|
+
y1 = (z - y0) - pio2_2t;
|
|
216
|
+
}
|
|
217
|
+
} else { // negative x
|
|
218
|
+
z = x + pio2_1;
|
|
219
|
+
if (ix != 0x3FF921FB) { // 33+53 bit pi is good enough
|
|
220
|
+
y0 = z + pio2_1t;
|
|
221
|
+
y1 = (z - y0) + pio2_1t;
|
|
222
|
+
} else { // near pi/2, use 33+33+53 bit pi
|
|
223
|
+
z += pio2_2;
|
|
224
|
+
y0 = z + pio2_2t;
|
|
225
|
+
y1 = (z - y0) + pio2_2t;
|
|
226
|
+
}
|
|
227
|
+
q = -1;
|
|
228
|
+
}
|
|
229
|
+
rempio2_y0 = y0;
|
|
230
|
+
rempio2_y1 = y1;
|
|
231
|
+
return q;
|
|
232
|
+
}
|
|
233
|
+
}
|
|
234
|
+
|
|
235
|
+
if (ix < 0x413921FB) { // |x| ~< 2^20*pi/2 (1647099)
|
|
236
|
+
// Use precise Cody Waite scheme
|
|
237
|
+
let q = nearest(x * invpio2);
|
|
238
|
+
let r = x - q * pio2_1;
|
|
239
|
+
let w = q * pio2_1t; // 1st round good to 85 bit
|
|
240
|
+
let j = ix >> 20;
|
|
241
|
+
let y0 = r - w;
|
|
242
|
+
let hi = <u32>(reinterpret<u64>(y0) >> 32);
|
|
243
|
+
let i = j - ((hi >> 20) & 0x7FF);
|
|
244
|
+
|
|
245
|
+
if (i > 16) { // 2nd iteration needed, good to 118
|
|
246
|
+
let t = r;
|
|
247
|
+
w = q * pio2_2;
|
|
248
|
+
r = t - w;
|
|
249
|
+
w = q * pio2_2t - ((t - r) - w);
|
|
250
|
+
y0 = r - w;
|
|
251
|
+
hi = <u32>(reinterpret<u64>(y0) >> 32);
|
|
252
|
+
i = j - ((hi >> 20) & 0x7FF);
|
|
253
|
+
if (i > 49) { // 3rd iteration need, 151 bits acc
|
|
254
|
+
let t = r;
|
|
255
|
+
w = q * pio2_3;
|
|
256
|
+
r = t - w;
|
|
257
|
+
w = q * pio2_3t - ((t - r) - w);
|
|
258
|
+
y0 = r - w;
|
|
259
|
+
}
|
|
260
|
+
}
|
|
261
|
+
let y1 = (r - y0) - w;
|
|
262
|
+
rempio2_y0 = y0;
|
|
263
|
+
rempio2_y1 = y1;
|
|
264
|
+
return <i32>q;
|
|
265
|
+
}
|
|
266
|
+
let q = pio2_large_quot(x, u);
|
|
267
|
+
return select(-q, q, sign);
|
|
268
|
+
}
|
|
269
|
+
|
|
270
|
+
/** @internal */
|
|
271
|
+
// @ts-ignore: decorator
|
|
272
|
+
@inline
|
|
273
|
+
function sin_kern(x: f64, y: f64, iy: i32): f64 { // see: musl/tree/src/math/__sin.c
|
|
274
|
+
const
|
|
275
|
+
S1 = reinterpret<f64>(0xBFC5555555555549), // -1.66666666666666324348e-01
|
|
276
|
+
S2 = reinterpret<f64>(0x3F8111111110F8A6), // 8.33333333332248946124e-03
|
|
277
|
+
S3 = reinterpret<f64>(0xBF2A01A019C161D5), // -1.98412698298579493134e-04
|
|
278
|
+
S4 = reinterpret<f64>(0x3EC71DE357B1FE7D), // 2.75573137070700676789e-06
|
|
279
|
+
S5 = reinterpret<f64>(0xBE5AE5E68A2B9CEB), // -2.50507602534068634195e-08
|
|
280
|
+
S6 = reinterpret<f64>(0x3DE5D93A5ACFD57C); // 1.58969099521155010221e-10
|
|
281
|
+
|
|
282
|
+
let z = x * x;
|
|
283
|
+
let w = z * z;
|
|
284
|
+
let r = S2 + z * (S3 + z * S4) + z * w * (S5 + z * S6);
|
|
285
|
+
let v = z * x;
|
|
286
|
+
if (!iy) {
|
|
287
|
+
return x + v * (S1 + z * r);
|
|
288
|
+
} else {
|
|
289
|
+
return x - ((z * (0.5 * y - v * r) - y) - v * S1);
|
|
290
|
+
}
|
|
291
|
+
}
|
|
292
|
+
|
|
293
|
+
/** @internal */
|
|
294
|
+
// @ts-ignore: decorator
|
|
295
|
+
@inline
|
|
296
|
+
function cos_kern(x: f64, y: f64): f64 { // see: musl/tree/src/math/__cos.c
|
|
297
|
+
const
|
|
298
|
+
C1 = reinterpret<f64>(0x3FA555555555554C), // 4.16666666666666019037e-02
|
|
299
|
+
C2 = reinterpret<f64>(0xBF56C16C16C15177), // -1.38888888888741095749e-03
|
|
300
|
+
C3 = reinterpret<f64>(0x3EFA01A019CB1590), // 2.48015872894767294178e-05
|
|
301
|
+
C4 = reinterpret<f64>(0xBE927E4F809C52AD), // -2.75573143513906633035e-07
|
|
302
|
+
C5 = reinterpret<f64>(0x3E21EE9EBDB4B1C4), // 2.08757232129817482790e-09
|
|
303
|
+
C6 = reinterpret<f64>(0xBDA8FAE9BE8838D4); // -1.13596475577881948265e-11
|
|
304
|
+
|
|
305
|
+
let z = x * x;
|
|
306
|
+
let w = z * z;
|
|
307
|
+
let r = z * (C1 + z * (C2 + z * C3)) + w * w * (C4 + z * (C5 + z * C6));
|
|
308
|
+
let hz = 0.5 * z;
|
|
309
|
+
w = 1.0 - hz;
|
|
310
|
+
return w + (((1.0 - w) - hz) + (z * r - x * y));
|
|
311
|
+
}
|
|
312
|
+
|
|
313
|
+
/** @internal */
|
|
314
|
+
function tan_kern(x: f64, y: f64, iy: i32): f64 { // see: src/lib/msun/src/k_tan.c
|
|
315
|
+
const
|
|
316
|
+
T0 = reinterpret<f64>(0x3FD5555555555563), // 3.33333333333334091986e-01
|
|
317
|
+
T1 = reinterpret<f64>(0x3FC111111110FE7A), // 1.33333333333201242699e-01
|
|
318
|
+
T2 = reinterpret<f64>(0x3FABA1BA1BB341FE), // 5.39682539762260521377e-02
|
|
319
|
+
T3 = reinterpret<f64>(0x3F9664F48406D637), // 2.18694882948595424599e-02
|
|
320
|
+
T4 = reinterpret<f64>(0x3F8226E3E96E8493), // 8.86323982359930005737e-03
|
|
321
|
+
T5 = reinterpret<f64>(0x3F6D6D22C9560328), // 3.59207910759131235356e-03
|
|
322
|
+
T6 = reinterpret<f64>(0x3F57DBC8FEE08315), // 1.45620945432529025516e-03
|
|
323
|
+
T7 = reinterpret<f64>(0x3F4344D8F2F26501), // 5.88041240820264096874e-04
|
|
324
|
+
T8 = reinterpret<f64>(0x3F3026F71A8D1068), // 2.46463134818469906812e-04
|
|
325
|
+
T9 = reinterpret<f64>(0x3F147E88A03792A6), // 7.81794442939557092300e-05
|
|
326
|
+
T10 = reinterpret<f64>(0x3F12B80F32F0A7E9), // 7.14072491382608190305e-05
|
|
327
|
+
T11 = reinterpret<f64>(0xBEF375CBDB605373), // -1.85586374855275456654e-05
|
|
328
|
+
T12 = reinterpret<f64>(0x3EFB2A7074BF7AD4); // 2.59073051863633712884e-05
|
|
329
|
+
|
|
330
|
+
const
|
|
331
|
+
one = reinterpret<f64>(0x3FF0000000000000), // 1.00000000000000000000e+00
|
|
332
|
+
pio4 = reinterpret<f64>(0x3FE921FB54442D18), // 7.85398163397448278999e-01
|
|
333
|
+
pio4lo = reinterpret<f64>(0x3C81A62633145C07); // 3.06161699786838301793e-17
|
|
334
|
+
|
|
335
|
+
let z: f64, r: f64, v: f64, w: f64, s: f64;
|
|
336
|
+
let hx = <i32>(reinterpret<u64>(x) >> 32); // high word of x
|
|
337
|
+
let ix = hx & 0x7FFFFFFF; // high word of |x|
|
|
338
|
+
let big = ix >= 0x3FE59428;
|
|
339
|
+
if (big) { // |x| >= 0.6744
|
|
340
|
+
if (hx < 0) { x = -x, y = -y; }
|
|
341
|
+
z = pio4 - x;
|
|
342
|
+
w = pio4lo - y;
|
|
343
|
+
x = z + w;
|
|
344
|
+
y = 0.0;
|
|
345
|
+
}
|
|
346
|
+
z = x * x;
|
|
347
|
+
w = z * z;
|
|
348
|
+
r = T1 + w * (T3 + w * (T5 + w * (T7 + w * (T9 + w * T11))));
|
|
349
|
+
v = z * (T2 + w * (T4 + w * (T6 + w * (T8 + w * (T10 + w * T12)))));
|
|
350
|
+
s = z * x;
|
|
351
|
+
r = y + z * (s * (r + v) + y);
|
|
352
|
+
r += T0 * s;
|
|
353
|
+
w = x + r;
|
|
354
|
+
if (big) {
|
|
355
|
+
v = iy;
|
|
356
|
+
return (1 - ((hx >> 30) & 2)) * (v - 2.0 * (x - (w * w / (w + v) - r)));
|
|
357
|
+
}
|
|
358
|
+
if (iy == 1) return w;
|
|
359
|
+
let a: f64, t: f64;
|
|
360
|
+
z = w;
|
|
361
|
+
z = reinterpret<f64>(reinterpret<u64>(z) & 0xFFFFFFFF00000000);
|
|
362
|
+
v = r - (z - x); // z + v = r + x
|
|
363
|
+
t = a = -one / w; // a = -1.0 / w
|
|
364
|
+
t = reinterpret<f64>(reinterpret<u64>(t) & 0xFFFFFFFF00000000);
|
|
365
|
+
s = one + t * z;
|
|
366
|
+
return t + a * (s + t * v);
|
|
367
|
+
}
|
|
368
|
+
|
|
369
|
+
/** @internal */
|
|
370
|
+
function dtoi32(x: f64): i32 {
|
|
371
|
+
if (ASC_SHRINK_LEVEL > 0) {
|
|
372
|
+
const inv32 = 1.0 / 4294967296;
|
|
373
|
+
return <i32><i64>(x - 4294967296 * floor(x * inv32));
|
|
374
|
+
} else {
|
|
375
|
+
let result = 0;
|
|
376
|
+
let u = reinterpret<u64>(x);
|
|
377
|
+
let e = (u >> 52) & 0x7FF;
|
|
378
|
+
if (e <= 1023 + 30) {
|
|
379
|
+
result = <i32>x;
|
|
380
|
+
} else if (e <= 1023 + 30 + 53) {
|
|
381
|
+
let v = (u & ((<u64>1 << 52) - 1)) | (<u64>1 << 52);
|
|
382
|
+
v = v << e - 1023 - 52 + 32;
|
|
383
|
+
result = <i32>(v >> 32);
|
|
384
|
+
result = select<i32>(-result, result, <i64>u < 0);
|
|
385
|
+
}
|
|
386
|
+
return result;
|
|
387
|
+
}
|
|
388
|
+
}
|
|
389
|
+
|
|
390
|
+
// @ts-ignore: decorator
|
|
391
|
+
@lazy let random_seeded = false;
|
|
392
|
+
|
|
393
|
+
// @ts-ignore: decorator
|
|
394
|
+
@lazy let random_state0_64: u64, random_state1_64: u64;
|
|
395
|
+
|
|
396
|
+
// @ts-ignore: decorator
|
|
397
|
+
@lazy let random_state0_32: u32, random_state1_32: u32;
|
|
398
|
+
|
|
399
|
+
function murmurHash3(h: u64): u64 { // Force all bits of a hash block to avalanche
|
|
400
|
+
h ^= h >> 33; // see: https://github.com/aappleby/smhasher
|
|
401
|
+
h *= 0xFF51AFD7ED558CCD;
|
|
402
|
+
h ^= h >> 33;
|
|
403
|
+
h *= 0xC4CEB9FE1A85EC53;
|
|
404
|
+
h ^= h >> 33;
|
|
405
|
+
return h;
|
|
406
|
+
}
|
|
407
|
+
|
|
408
|
+
function splitMix32(h: u32): u32 {
|
|
409
|
+
h += 0x6D2B79F5;
|
|
410
|
+
h = (h ^ (h >> 15)) * (h | 1);
|
|
411
|
+
h ^= h + (h ^ (h >> 7)) * (h | 61);
|
|
412
|
+
return h ^ (h >> 14);
|
|
413
|
+
}
|
|
414
|
+
|
|
415
|
+
export namespace NativeMath {
|
|
416
|
+
|
|
417
|
+
// @ts-ignore: decorator
|
|
418
|
+
@lazy
|
|
419
|
+
export const E = reinterpret<f64>(0x4005BF0A8B145769); // 2.7182818284590452354
|
|
420
|
+
|
|
421
|
+
// @ts-ignore: decorator
|
|
422
|
+
@lazy
|
|
423
|
+
export const LN2 = reinterpret<f64>(0x3FE62E42FEFA39EF); // 0.69314718055994530942
|
|
424
|
+
|
|
425
|
+
// @ts-ignore: decorator
|
|
426
|
+
@lazy
|
|
427
|
+
export const LN10 = reinterpret<f64>(0x40026BB1BBB55516); // 2.30258509299404568402
|
|
428
|
+
|
|
429
|
+
// @ts-ignore: decorator
|
|
430
|
+
@lazy
|
|
431
|
+
export const LOG2E = reinterpret<f64>(0x3FF71547652B82FE); // 1.4426950408889634074
|
|
432
|
+
|
|
433
|
+
// @ts-ignore: decorator
|
|
434
|
+
@lazy
|
|
435
|
+
export const LOG10E = reinterpret<f64>(0x3FDBCB7B1526E50E); // 0.43429448190325182765
|
|
436
|
+
|
|
437
|
+
// @ts-ignore: decorator
|
|
438
|
+
@lazy
|
|
439
|
+
export const PI = reinterpret<f64>(0x400921FB54442D18); // 3.14159265358979323846
|
|
440
|
+
|
|
441
|
+
// @ts-ignore: decorator
|
|
442
|
+
@lazy
|
|
443
|
+
export const SQRT1_2 = reinterpret<f64>(0x3FE6A09E667F3BCD); // 0.70710678118654752440
|
|
444
|
+
|
|
445
|
+
// @ts-ignore: decorator
|
|
446
|
+
@lazy
|
|
447
|
+
export const SQRT2 = reinterpret<f64>(0x3FF6A09E667F3BCD); // 1.41421356237309504880
|
|
448
|
+
|
|
449
|
+
// @ts-ignore: decorator
|
|
450
|
+
@lazy
|
|
451
|
+
export let sincos_sin: f64 = 0;
|
|
452
|
+
|
|
453
|
+
// @ts-ignore: decorator
|
|
454
|
+
@lazy
|
|
455
|
+
export let sincos_cos: f64 = 0;
|
|
456
|
+
|
|
457
|
+
// @ts-ignore: decorator
|
|
458
|
+
@inline export function abs(x: f64): f64 {
|
|
459
|
+
return builtin_abs<f64>(x);
|
|
460
|
+
}
|
|
461
|
+
|
|
462
|
+
export function acos(x: f64): f64 { // see: musl/src/math/acos.c and SUN COPYRIGHT NOTICE above
|
|
463
|
+
const
|
|
464
|
+
pio2_hi = reinterpret<f64>(0x3FF921FB54442D18), // 1.57079632679489655800e+00
|
|
465
|
+
pio2_lo = reinterpret<f64>(0x3C91A62633145C07), // 6.12323399573676603587e-17
|
|
466
|
+
Ox1p_120f = reinterpret<f32>(0x03800000);
|
|
467
|
+
|
|
468
|
+
let hx = <u32>(reinterpret<u64>(x) >> 32);
|
|
469
|
+
let ix = hx & 0x7FFFFFFF;
|
|
470
|
+
if (ix >= 0x3FF00000) {
|
|
471
|
+
let lx = <u32>reinterpret<u64>(x);
|
|
472
|
+
if ((ix - 0x3FF00000 | lx) == 0) {
|
|
473
|
+
if (<i32>hx < 0) return 2 * pio2_hi + Ox1p_120f;
|
|
474
|
+
return 0;
|
|
475
|
+
}
|
|
476
|
+
return 0 / (x - x);
|
|
477
|
+
}
|
|
478
|
+
if (ix < 0x3FE00000) {
|
|
479
|
+
if (ix <= 0x3C600000) return pio2_hi + Ox1p_120f;
|
|
480
|
+
return pio2_hi - (x - (pio2_lo - x * R(x * x)));
|
|
481
|
+
}
|
|
482
|
+
let s: f64, w: f64, z: f64;
|
|
483
|
+
if (<i32>hx < 0) {
|
|
484
|
+
// z = (1.0 + x) * 0.5;
|
|
485
|
+
z = 0.5 + x * 0.5;
|
|
486
|
+
s = builtin_sqrt<f64>(z);
|
|
487
|
+
w = R(z) * s - pio2_lo;
|
|
488
|
+
return 2 * (pio2_hi - (s + w));
|
|
489
|
+
}
|
|
490
|
+
// z = (1.0 - x) * 0.5;
|
|
491
|
+
z = 0.5 - x * 0.5;
|
|
492
|
+
s = builtin_sqrt<f64>(z);
|
|
493
|
+
let df = reinterpret<f64>(reinterpret<u64>(s) & 0xFFFFFFFF00000000);
|
|
494
|
+
let c = (z - df * df) / (s + df);
|
|
495
|
+
w = R(z) * s + c;
|
|
496
|
+
return 2 * (df + w);
|
|
497
|
+
}
|
|
498
|
+
|
|
499
|
+
export function acosh(x: f64): f64 { // see: musl/src/math/acosh.c
|
|
500
|
+
const s = reinterpret<f64>(0x3FE62E42FEFA39EF);
|
|
501
|
+
let u = reinterpret<u64>(x);
|
|
502
|
+
// Prevent propagation for all input values less than 1.0.
|
|
503
|
+
// Note musl lib didn't fix this yet.
|
|
504
|
+
if (<i64>u < 0x3FF0000000000000) return (x - x) / 0.0;
|
|
505
|
+
let e = u >> 52 & 0x7FF;
|
|
506
|
+
if (e < 0x3FF + 1) return log1p(x - 1 + builtin_sqrt<f64>((x - 1) * (x - 1) + 2 * (x - 1)));
|
|
507
|
+
if (e < 0x3FF + 26) return log(2 * x - 1 / (x + builtin_sqrt<f64>(x * x - 1)));
|
|
508
|
+
return log(x) + s;
|
|
509
|
+
}
|
|
510
|
+
|
|
511
|
+
export function asin(x: f64): f64 { // see: musl/src/math/asin.c and SUN COPYRIGHT NOTICE above
|
|
512
|
+
const
|
|
513
|
+
pio2_hi = reinterpret<f64>(0x3FF921FB54442D18), // 1.57079632679489655800e+00
|
|
514
|
+
pio2_lo = reinterpret<f64>(0x3C91A62633145C07), // 6.12323399573676603587e-17
|
|
515
|
+
Ox1p_120f = reinterpret<f32>(0x03800000);
|
|
516
|
+
|
|
517
|
+
let hx = <u32>(reinterpret<u64>(x) >> 32);
|
|
518
|
+
let ix = hx & 0x7FFFFFFF;
|
|
519
|
+
if (ix >= 0x3FF00000) {
|
|
520
|
+
let lx = <u32>reinterpret<u64>(x);
|
|
521
|
+
if ((ix - 0x3FF00000 | lx) == 0) return x * pio2_hi + Ox1p_120f;
|
|
522
|
+
return 0 / (x - x);
|
|
523
|
+
}
|
|
524
|
+
if (ix < 0x3FE00000) {
|
|
525
|
+
if (ix < 0x3E500000 && ix >= 0x00100000) return x;
|
|
526
|
+
return x + x * R(x * x);
|
|
527
|
+
}
|
|
528
|
+
// let z = (1.0 - builtin_abs<f64>(x)) * 0.5;
|
|
529
|
+
let z = 0.5 - builtin_abs<f64>(x) * 0.5;
|
|
530
|
+
let s = builtin_sqrt<f64>(z);
|
|
531
|
+
let r = R(z);
|
|
532
|
+
if (ix >= 0x3FEF3333) x = pio2_hi - (2 * (s + s * r) - pio2_lo);
|
|
533
|
+
else {
|
|
534
|
+
let f = reinterpret<f64>(reinterpret<u64>(s) & 0xFFFFFFFF00000000);
|
|
535
|
+
let c = (z - f * f) / (s + f);
|
|
536
|
+
x = 0.5 * pio2_hi - (2 * s * r - (pio2_lo - 2 * c) - (0.5 * pio2_hi - 2 * f));
|
|
537
|
+
}
|
|
538
|
+
return select(-x, x, <i32>hx < 0);
|
|
539
|
+
}
|
|
540
|
+
|
|
541
|
+
export function asinh(x: f64): f64 { // see: musl/src/math/asinh.c
|
|
542
|
+
const c = reinterpret<f64>(0x3FE62E42FEFA39EF); // 0.693147180559945309417232121458176568
|
|
543
|
+
let u = reinterpret<u64>(x);
|
|
544
|
+
let e = u >> 52 & 0x7FF;
|
|
545
|
+
let y = reinterpret<f64>(u & 0x7FFFFFFFFFFFFFFF);
|
|
546
|
+
if (e >= 0x3FF + 26) y = log(y) + c;
|
|
547
|
+
else if (e >= 0x3FF + 1) y = log(2 * y + 1 / (builtin_sqrt<f64>(y * y + 1) + y));
|
|
548
|
+
else if (e >= 0x3FF - 26) y = log1p(y + y * y / (builtin_sqrt<f64>(y * y + 1) + 1));
|
|
549
|
+
return builtin_copysign(y, x);
|
|
550
|
+
}
|
|
551
|
+
|
|
552
|
+
export function atan(x: f64): f64 { // see musl/src/math/atan.c and SUN COPYRIGHT NOTICE above
|
|
553
|
+
const
|
|
554
|
+
atanhi0 = reinterpret<f64>(0x3FDDAC670561BB4F), // 4.63647609000806093515e-01
|
|
555
|
+
atanhi1 = reinterpret<f64>(0x3FE921FB54442D18), // 7.85398163397448278999e-01
|
|
556
|
+
atanhi2 = reinterpret<f64>(0x3FEF730BD281F69B), // 9.82793723247329054082e-01
|
|
557
|
+
atanhi3 = reinterpret<f64>(0x3FF921FB54442D18), // 1.57079632679489655800e+00
|
|
558
|
+
atanlo0 = reinterpret<f64>(0x3C7A2B7F222F65E2), // 2.26987774529616870924e-17
|
|
559
|
+
atanlo1 = reinterpret<f64>(0x3C81A62633145C07), // 3.06161699786838301793e-17
|
|
560
|
+
atanlo2 = reinterpret<f64>(0x3C7007887AF0CBBD), // 1.39033110312309984516e-17
|
|
561
|
+
atanlo3 = reinterpret<f64>(0x3C91A62633145C07), // 6.12323399573676603587e-17
|
|
562
|
+
aT0 = reinterpret<f64>(0x3FD555555555550D), // 3.33333333333329318027e-01
|
|
563
|
+
aT1 = reinterpret<f64>(0xBFC999999998EBC4), // -1.99999999998764832476e-01
|
|
564
|
+
aT2 = reinterpret<f64>(0x3FC24924920083FF), // 1.42857142725034663711e-01
|
|
565
|
+
aT3 = reinterpret<f64>(0xBFBC71C6FE231671), // -1.11111104054623557880e-01,
|
|
566
|
+
aT4 = reinterpret<f64>(0x3FB745CDC54C206E), // 9.09088713343650656196e-02
|
|
567
|
+
aT5 = reinterpret<f64>(0xBFB3B0F2AF749A6D), // -7.69187620504482999495e-02
|
|
568
|
+
aT6 = reinterpret<f64>(0x3FB10D66A0D03D51), // 6.66107313738753120669e-02
|
|
569
|
+
aT7 = reinterpret<f64>(0xBFADDE2D52DEFD9A), // -5.83357013379057348645e-02
|
|
570
|
+
aT8 = reinterpret<f64>(0x3FA97B4B24760DEB), // 4.97687799461593236017e-02
|
|
571
|
+
aT9 = reinterpret<f64>(0xBFA2B4442C6A6C2F), // -3.65315727442169155270e-02
|
|
572
|
+
aT10 = reinterpret<f64>(0x3F90AD3AE322DA11), // 1.62858201153657823623e-02
|
|
573
|
+
Ox1p_120f = reinterpret<f32>(0x03800000);
|
|
574
|
+
|
|
575
|
+
let ix = <u32>(reinterpret<u64>(x) >> 32);
|
|
576
|
+
let sx = x;
|
|
577
|
+
ix &= 0x7FFFFFFF;
|
|
578
|
+
let z: f64;
|
|
579
|
+
if (ix >= 0x44100000) {
|
|
580
|
+
if (isNaN(x)) return x;
|
|
581
|
+
z = atanhi3 + Ox1p_120f;
|
|
582
|
+
return builtin_copysign<f64>(z, sx);
|
|
583
|
+
}
|
|
584
|
+
let id: i32;
|
|
585
|
+
if (ix < 0x3FDC0000) {
|
|
586
|
+
if (ix < 0x3E400000) return x;
|
|
587
|
+
id = -1;
|
|
588
|
+
} else {
|
|
589
|
+
x = builtin_abs<f64>(x);
|
|
590
|
+
if (ix < 0x3FF30000) {
|
|
591
|
+
if (ix < 0x3FE60000) {
|
|
592
|
+
id = 0;
|
|
593
|
+
x = (2.0 * x - 1.0) / (2.0 + x);
|
|
594
|
+
} else {
|
|
595
|
+
id = 1;
|
|
596
|
+
x = (x - 1.0) / (x + 1.0);
|
|
597
|
+
}
|
|
598
|
+
} else {
|
|
599
|
+
if (ix < 0x40038000) {
|
|
600
|
+
id = 2;
|
|
601
|
+
x = (x - 1.5) / (1.0 + 1.5 * x);
|
|
602
|
+
} else {
|
|
603
|
+
id = 3;
|
|
604
|
+
x = -1.0 / x;
|
|
605
|
+
}
|
|
606
|
+
}
|
|
607
|
+
}
|
|
608
|
+
z = x * x;
|
|
609
|
+
let w = z * z;
|
|
610
|
+
let s1 = z * (aT0 + w * (aT2 + w * (aT4 + w * (aT6 + w * (aT8 + w * aT10)))));
|
|
611
|
+
let s2 = w * (aT1 + w * (aT3 + w * (aT5 + w * (aT7 + w * aT9))));
|
|
612
|
+
let s3 = x * (s1 + s2);
|
|
613
|
+
if (id < 0) return x - s3;
|
|
614
|
+
switch (id) {
|
|
615
|
+
case 0: { z = atanhi0 - ((s3 - atanlo0) - x); break; }
|
|
616
|
+
case 1: { z = atanhi1 - ((s3 - atanlo1) - x); break; }
|
|
617
|
+
case 2: { z = atanhi2 - ((s3 - atanlo2) - x); break; }
|
|
618
|
+
case 3: { z = atanhi3 - ((s3 - atanlo3) - x); break; }
|
|
619
|
+
default: unreachable();
|
|
620
|
+
}
|
|
621
|
+
return builtin_copysign<f64>(z, sx);
|
|
622
|
+
}
|
|
623
|
+
|
|
624
|
+
export function atanh(x: f64): f64 { // see: musl/src/math/atanh.c
|
|
625
|
+
let u = reinterpret<u64>(x);
|
|
626
|
+
let e = u >> 52 & 0x7FF;
|
|
627
|
+
let y = builtin_abs(x);
|
|
628
|
+
if (e < 0x3FF - 1) {
|
|
629
|
+
if (e >= 0x3FF - 32) y = 0.5 * log1p(2 * y + 2 * y * y / (1 - y));
|
|
630
|
+
} else {
|
|
631
|
+
y = 0.5 * log1p(2 * (y / (1 - y)));
|
|
632
|
+
}
|
|
633
|
+
return builtin_copysign<f64>(y, x);
|
|
634
|
+
}
|
|
635
|
+
|
|
636
|
+
export function atan2(y: f64, x: f64): f64 { // see: musl/src/math/atan2.c and SUN COPYRIGHT NOTICE above
|
|
637
|
+
const pi_lo = reinterpret<f64>(0x3CA1A62633145C07); // 1.2246467991473531772E-16
|
|
638
|
+
if (isNaN(x) || isNaN(y)) return x + y;
|
|
639
|
+
let u = reinterpret<u64>(x);
|
|
640
|
+
let ix = <u32>(u >> 32);
|
|
641
|
+
let lx = <u32>u;
|
|
642
|
+
u = reinterpret<u64>(y);
|
|
643
|
+
let iy = <u32>(u >> 32);
|
|
644
|
+
let ly = <u32>u;
|
|
645
|
+
if ((ix - 0x3FF00000 | lx) == 0) return atan(y);
|
|
646
|
+
let m = ((iy >> 31) & 1) | ((ix >> 30) & 2);
|
|
647
|
+
ix = ix & 0x7FFFFFFF;
|
|
648
|
+
iy = iy & 0x7FFFFFFF;
|
|
649
|
+
if ((iy | ly) == 0) {
|
|
650
|
+
switch (m) {
|
|
651
|
+
case 0:
|
|
652
|
+
case 1: return y;
|
|
653
|
+
case 2: return PI;
|
|
654
|
+
case 3: return -PI;
|
|
655
|
+
}
|
|
656
|
+
}
|
|
657
|
+
if ((ix | lx) == 0) return m & 1 ? -PI / 2 : PI / 2;
|
|
658
|
+
if (ix == 0x7FF00000) {
|
|
659
|
+
if (iy == 0x7FF00000) {
|
|
660
|
+
let t = m & 2 ? 3 * PI / 4 : PI / 4;
|
|
661
|
+
return m & 1 ? -t : t;
|
|
662
|
+
} else {
|
|
663
|
+
let t = m & 2 ? PI : 0;
|
|
664
|
+
return m & 1 ? -t : t;
|
|
665
|
+
}
|
|
666
|
+
}
|
|
667
|
+
let z: f64;
|
|
668
|
+
if (ix + (64 << 20) < iy || iy == 0x7FF00000) return m & 1 ? -PI / 2 : PI / 2;
|
|
669
|
+
if ((m & 2) && iy + (64 << 20) < ix) z = 0;
|
|
670
|
+
else z = atan(builtin_abs<f64>(y / x));
|
|
671
|
+
switch (m) {
|
|
672
|
+
case 0: return z;
|
|
673
|
+
case 1: return -z;
|
|
674
|
+
case 2: return PI - (z - pi_lo);
|
|
675
|
+
case 3: return (z - pi_lo) - PI;
|
|
676
|
+
}
|
|
677
|
+
unreachable();
|
|
678
|
+
return 0;
|
|
679
|
+
}
|
|
680
|
+
|
|
681
|
+
export function cbrt(x: f64): f64 { // see: musl/src/math/cbrt.c and SUN COPYRIGHT NOTICE above
|
|
682
|
+
const
|
|
683
|
+
B1 = <u32>715094163,
|
|
684
|
+
B2 = <u32>696219795,
|
|
685
|
+
P0 = reinterpret<f64>(0x3FFE03E60F61E692), // 1.87595182427177009643
|
|
686
|
+
P1 = reinterpret<f64>(0xBFFE28E092F02420), // -1.88497979543377169875
|
|
687
|
+
P2 = reinterpret<f64>(0x3FF9F1604A49D6C2), // 1.621429720105354466140
|
|
688
|
+
P3 = reinterpret<f64>(0xBFE844CBBEE751D9), // -0.758397934778766047437
|
|
689
|
+
P4 = reinterpret<f64>(0x3FC2B000D4E4EDD7), // 0.145996192886612446982
|
|
690
|
+
Ox1p54 = reinterpret<f64>(0x4350000000000000); // 0x1p54
|
|
691
|
+
|
|
692
|
+
let u = reinterpret<u64>(x);
|
|
693
|
+
let hx = <u32>(u >> 32) & 0x7FFFFFFF;
|
|
694
|
+
if (hx >= 0x7FF00000) return x + x;
|
|
695
|
+
if (hx < 0x00100000) {
|
|
696
|
+
u = reinterpret<u64>(x * Ox1p54);
|
|
697
|
+
hx = <u32>(u >> 32) & 0x7FFFFFFF;
|
|
698
|
+
if (hx == 0) return x;
|
|
699
|
+
hx = hx / 3 + B2;
|
|
700
|
+
} else {
|
|
701
|
+
hx = hx / 3 + B1;
|
|
702
|
+
}
|
|
703
|
+
u &= 1 << 63;
|
|
704
|
+
u |= <u64>hx << 32;
|
|
705
|
+
let t = reinterpret<f64>(u);
|
|
706
|
+
let r = (t * t) * (t / x);
|
|
707
|
+
t = t * ((P0 + r * (P1 + r * P2)) + ((r * r) * r) * (P3 + r * P4));
|
|
708
|
+
t = reinterpret<f64>((reinterpret<u64>(t) + 0x80000000) & 0xFFFFFFFFC0000000);
|
|
709
|
+
let s = t * t;
|
|
710
|
+
r = x / s;
|
|
711
|
+
r = (r - t) / (2 * t + r);
|
|
712
|
+
t = t + t * r;
|
|
713
|
+
return t;
|
|
714
|
+
}
|
|
715
|
+
|
|
716
|
+
// @ts-ignore: decorator
|
|
717
|
+
@inline
|
|
718
|
+
export function ceil(x: f64): f64 {
|
|
719
|
+
return builtin_ceil<f64>(x);
|
|
720
|
+
}
|
|
721
|
+
|
|
722
|
+
export function clz32(x: f64): f64 {
|
|
723
|
+
if (!isFinite(x)) return 32;
|
|
724
|
+
/*
|
|
725
|
+
* Wasm (MVP) and JS have different approaches for double->int conversions.
|
|
726
|
+
*
|
|
727
|
+
* For emulate JS conversion behavior and avoid trapping from wasm we should modulate by MAX_INT
|
|
728
|
+
* our float-point arguments before actual convertion to integers.
|
|
729
|
+
*/
|
|
730
|
+
return builtin_clz(dtoi32(x));
|
|
731
|
+
}
|
|
732
|
+
|
|
733
|
+
export function cos(x: f64): f64 { // see: musl/src/math/cos.c
|
|
734
|
+
let u = reinterpret<u64>(x);
|
|
735
|
+
let ux = u32(u >> 32);
|
|
736
|
+
let sign = ux >> 31;
|
|
737
|
+
|
|
738
|
+
ux &= 0x7FFFFFFF;
|
|
739
|
+
|
|
740
|
+
// |x| ~< pi/4
|
|
741
|
+
if (ux <= 0x3FE921FB) {
|
|
742
|
+
if (ux < 0x3E46A09E) { // |x| < 2**-27 * sqrt(2)
|
|
743
|
+
return 1.0;
|
|
744
|
+
}
|
|
745
|
+
return cos_kern(x, 0);
|
|
746
|
+
}
|
|
747
|
+
|
|
748
|
+
// sin(Inf or NaN) is NaN
|
|
749
|
+
if (ux >= 0x7FF00000) return x - x;
|
|
750
|
+
|
|
751
|
+
// argument reduction needed
|
|
752
|
+
let n = rempio2(x, u, sign);
|
|
753
|
+
let y0 = rempio2_y0;
|
|
754
|
+
let y1 = rempio2_y1;
|
|
755
|
+
|
|
756
|
+
x = n & 1 ? sin_kern(y0, y1, 1) : cos_kern(y0, y1);
|
|
757
|
+
return (n + 1) & 2 ? -x : x;
|
|
758
|
+
}
|
|
759
|
+
|
|
760
|
+
export function cosh(x: f64): f64 { // see: musl/src/math/cosh.c
|
|
761
|
+
let u = reinterpret<u64>(x);
|
|
762
|
+
u &= 0x7FFFFFFFFFFFFFFF;
|
|
763
|
+
x = reinterpret<f64>(u);
|
|
764
|
+
let w = <u32>(u >> 32);
|
|
765
|
+
let t: f64;
|
|
766
|
+
if (w < 0x3FE62E42) {
|
|
767
|
+
if (w < 0x3FF00000 - (26 << 20)) return 1;
|
|
768
|
+
t = expm1(x);
|
|
769
|
+
// return 1 + t * t / (2 * (1 + t));
|
|
770
|
+
return 1 + t * t / (2 + 2 * t);
|
|
771
|
+
}
|
|
772
|
+
if (w < 0x40862E42) {
|
|
773
|
+
t = exp(x);
|
|
774
|
+
return 0.5 * (t + 1 / t);
|
|
775
|
+
}
|
|
776
|
+
t = expo2(x, 1);
|
|
777
|
+
return t;
|
|
778
|
+
}
|
|
779
|
+
|
|
780
|
+
export function exp(x: f64): f64 { // see: musl/src/math/exp.c and SUN COPYRIGHT NOTICE above
|
|
781
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
782
|
+
return exp_lut(x);
|
|
783
|
+
} else {
|
|
784
|
+
const
|
|
785
|
+
ln2hi = reinterpret<f64>(0x3FE62E42FEE00000), // 6.93147180369123816490e-01
|
|
786
|
+
ln2lo = reinterpret<f64>(0x3DEA39EF35793C76), // 1.90821492927058770002e-10
|
|
787
|
+
invln2 = reinterpret<f64>(0x3FF71547652B82FE), // 1.44269504088896338700e+00
|
|
788
|
+
P1 = reinterpret<f64>(0x3FC555555555553E), // 1.66666666666666019037e-01
|
|
789
|
+
P2 = reinterpret<f64>(0xBF66C16C16BEBD93), // -2.77777777770155933842e-03
|
|
790
|
+
P3 = reinterpret<f64>(0x3F11566AAF25DE2C), // 6.61375632143793436117e-05
|
|
791
|
+
P4 = reinterpret<f64>(0xBEBBBD41C5D26BF1), // -1.65339022054652515390e-06
|
|
792
|
+
P5 = reinterpret<f64>(0x3E66376972BEA4D0), // 4.13813679705723846039e-08
|
|
793
|
+
overflow = reinterpret<f64>(0x40862E42FEFA39EF), // 709.782712893383973096
|
|
794
|
+
underflow = reinterpret<f64>(0xC0874910D52D3051), // -745.13321910194110842
|
|
795
|
+
Ox1p1023 = reinterpret<f64>(0x7FE0000000000000); // 0x1p1023
|
|
796
|
+
|
|
797
|
+
let hx = u32(reinterpret<u64>(x) >> 32);
|
|
798
|
+
let sign = hx >> 31;
|
|
799
|
+
hx &= 0x7FFFFFFF;
|
|
800
|
+
if (hx >= 0x4086232B) {
|
|
801
|
+
if (isNaN(x)) return x;
|
|
802
|
+
if (x > overflow) return x * Ox1p1023;
|
|
803
|
+
if (x < underflow) return 0;
|
|
804
|
+
}
|
|
805
|
+
let hi: f64, lo: f64 = 0;
|
|
806
|
+
let k = 0;
|
|
807
|
+
if (hx > 0x3FD62E42) {
|
|
808
|
+
if (hx >= 0x3FF0A2B2) {
|
|
809
|
+
k = i32(invln2 * x + builtin_copysign<f64>(0.5, x));
|
|
810
|
+
} else {
|
|
811
|
+
k = 1 - (sign << 1);
|
|
812
|
+
}
|
|
813
|
+
hi = x - k * ln2hi;
|
|
814
|
+
lo = k * ln2lo;
|
|
815
|
+
x = hi - lo;
|
|
816
|
+
} else if (hx > 0x3E300000) {
|
|
817
|
+
hi = x;
|
|
818
|
+
} else return 1.0 + x;
|
|
819
|
+
let xs = x * x;
|
|
820
|
+
// let c = x - xp2 * (P1 + xp2 * (P2 + xp2 * (P3 + xp2 * (P4 + xp2 * P5))));
|
|
821
|
+
let xq = xs * xs;
|
|
822
|
+
let c = x - (xs * P1 + xq * ((P2 + xs * P3) + xq * (P4 + xs * P5)));
|
|
823
|
+
let y = 1.0 + (x * c / (2 - c) - lo + hi);
|
|
824
|
+
return k == 0 ? y : scalbn(y, k);
|
|
825
|
+
}
|
|
826
|
+
}
|
|
827
|
+
|
|
828
|
+
export function exp2(x: f64): f64 {
|
|
829
|
+
return exp2_lut(x);
|
|
830
|
+
}
|
|
831
|
+
|
|
832
|
+
export function expm1(x: f64): f64 { // see: musl/src/math/expm1.c and SUN COPYRIGHT NOTICE above
|
|
833
|
+
const
|
|
834
|
+
o_threshold = reinterpret<f64>(0x40862E42FEFA39EF), // 7.09782712893383973096e+02
|
|
835
|
+
ln2_hi = reinterpret<f64>(0x3FE62E42FEE00000), // 6.93147180369123816490e-01
|
|
836
|
+
ln2_lo = reinterpret<f64>(0x3DEA39EF35793C76), // 1.90821492927058770002e-10
|
|
837
|
+
invln2 = reinterpret<f64>(0x3FF71547652B82FE), // 1.44269504088896338700e+00
|
|
838
|
+
Q1 = reinterpret<f64>(0xBFA11111111110F4), // -3.33333333333331316428e-02
|
|
839
|
+
Q2 = reinterpret<f64>(0x3F5A01A019FE5585), // 1.58730158725481460165e-03
|
|
840
|
+
Q3 = reinterpret<f64>(0xBF14CE199EAADBB7), // -7.93650757867487942473e-05
|
|
841
|
+
Q4 = reinterpret<f64>(0x3ED0CFCA86E65239), // 4.00821782732936239552e-06
|
|
842
|
+
Q5 = reinterpret<f64>(0xBE8AFDB76E09C32D), // -2.01099218183624371326e-07
|
|
843
|
+
Ox1p1023 = reinterpret<f64>(0x7FE0000000000000); // 0x1p1023
|
|
844
|
+
|
|
845
|
+
let u = reinterpret<u64>(x);
|
|
846
|
+
let hx = u32(u >> 32) & 0x7FFFFFFF;
|
|
847
|
+
let sign = u32(u >> 63);
|
|
848
|
+
let k = 0;
|
|
849
|
+
if (hx >= 0x4043687A) {
|
|
850
|
+
if (isNaN(x)) return x;
|
|
851
|
+
if (sign) return -1;
|
|
852
|
+
if (x > o_threshold) return x * Ox1p1023;
|
|
853
|
+
}
|
|
854
|
+
let c = 0.0, t: f64;
|
|
855
|
+
if (hx > 0x3FD62E42) {
|
|
856
|
+
k = select<i32>(
|
|
857
|
+
1 - (sign << 1),
|
|
858
|
+
i32(invln2 * x + builtin_copysign<f64>(0.5, x)),
|
|
859
|
+
hx < 0x3FF0A2B2
|
|
860
|
+
);
|
|
861
|
+
t = <f64>k;
|
|
862
|
+
let hi = x - t * ln2_hi;
|
|
863
|
+
let lo = t * ln2_lo;
|
|
864
|
+
x = hi - lo;
|
|
865
|
+
c = (hi - x) - lo;
|
|
866
|
+
} else if (hx < 0x3C900000) return x;
|
|
867
|
+
let hfx = 0.5 * x;
|
|
868
|
+
let hxs = x * hfx;
|
|
869
|
+
// let r1 = 1.0 + hxs * (Q1 + hxs * (Q2 + hxs * (Q3 + hxs * (Q4 + hxs * Q5))));
|
|
870
|
+
let hxq = hxs * hxs;
|
|
871
|
+
let r1 = (1.0 + hxs * Q1) + hxq * ((Q2 + hxs * Q3) + hxq * (Q4 + hxs * Q5));
|
|
872
|
+
t = 3.0 - r1 * hfx;
|
|
873
|
+
let e = hxs * ((r1 - t) / (6.0 - x * t));
|
|
874
|
+
if (k == 0) return x - (x * e - hxs);
|
|
875
|
+
e = x * (e - c) - c;
|
|
876
|
+
e -= hxs;
|
|
877
|
+
if (k == -1) return 0.5 * (x - e) - 0.5;
|
|
878
|
+
if (k == 1) {
|
|
879
|
+
if (x < -0.25) return -2.0 * (e - (x + 0.5));
|
|
880
|
+
return 1.0 + 2.0 * (x - e);
|
|
881
|
+
}
|
|
882
|
+
u = (0x3FF + k) << 52;
|
|
883
|
+
let twopk = reinterpret<f64>(u);
|
|
884
|
+
let y: f64;
|
|
885
|
+
if (k < 0 || k > 56) {
|
|
886
|
+
y = x - e + 1.0;
|
|
887
|
+
if (k == 1024) y = y * 2.0 * Ox1p1023;
|
|
888
|
+
else y = y * twopk;
|
|
889
|
+
return y - 1.0;
|
|
890
|
+
}
|
|
891
|
+
u = (0x3FF - k) << 52;
|
|
892
|
+
y = reinterpret<f64>(u);
|
|
893
|
+
if (k < 20) y = (1 - y) - e;
|
|
894
|
+
else y = 1 - (e + y);
|
|
895
|
+
return (x + y) * twopk;
|
|
896
|
+
}
|
|
897
|
+
|
|
898
|
+
// @ts-ignore: decorator
|
|
899
|
+
@inline
|
|
900
|
+
export function floor(x: f64): f64 {
|
|
901
|
+
return builtin_floor<f64>(x);
|
|
902
|
+
}
|
|
903
|
+
|
|
904
|
+
// @ts-ignore: decorator
|
|
905
|
+
@inline
|
|
906
|
+
export function fround(x: f64): f64 {
|
|
907
|
+
return <f32>x;
|
|
908
|
+
}
|
|
909
|
+
|
|
910
|
+
export function hypot(x: f64, y: f64): f64 { // see: musl/src/math/hypot.c
|
|
911
|
+
const
|
|
912
|
+
SPLIT = reinterpret<f64>(0x41A0000000000000) + 1, // 0x1p27 + 1
|
|
913
|
+
Ox1p700 = reinterpret<f64>(0x6BB0000000000000),
|
|
914
|
+
Ox1p_700 = reinterpret<f64>(0x1430000000000000);
|
|
915
|
+
|
|
916
|
+
let ux = reinterpret<u64>(x);
|
|
917
|
+
let uy = reinterpret<u64>(y);
|
|
918
|
+
ux &= 0x7FFFFFFFFFFFFFFF;
|
|
919
|
+
uy &= 0x7FFFFFFFFFFFFFFF;
|
|
920
|
+
if (ux < uy) {
|
|
921
|
+
let ut = ux;
|
|
922
|
+
ux = uy;
|
|
923
|
+
uy = ut;
|
|
924
|
+
}
|
|
925
|
+
let ex = i32(ux >> 52);
|
|
926
|
+
let ey = i32(uy >> 52);
|
|
927
|
+
y = reinterpret<f64>(uy);
|
|
928
|
+
if (ey == 0x7FF) return y;
|
|
929
|
+
x = reinterpret<f64>(ux);
|
|
930
|
+
if (ex == 0x7FF || uy == 0) return x;
|
|
931
|
+
if (ex - ey > 64) return x + y;
|
|
932
|
+
let z = 1.0;
|
|
933
|
+
if (ex > 0x3FF + 510) {
|
|
934
|
+
z = Ox1p700;
|
|
935
|
+
x *= Ox1p_700;
|
|
936
|
+
y *= Ox1p_700;
|
|
937
|
+
} else if (ey < 0x3FF - 450) {
|
|
938
|
+
z = Ox1p_700;
|
|
939
|
+
x *= Ox1p700;
|
|
940
|
+
y *= Ox1p700;
|
|
941
|
+
}
|
|
942
|
+
let c = x * SPLIT;
|
|
943
|
+
let h = x - c + c;
|
|
944
|
+
let l = x - h;
|
|
945
|
+
let hx = x * x;
|
|
946
|
+
let lx = h * h - hx + (2 * h + l) * l;
|
|
947
|
+
c = y * SPLIT;
|
|
948
|
+
h = y - c + c;
|
|
949
|
+
l = y - h;
|
|
950
|
+
let hy = y * y;
|
|
951
|
+
let ly = h * h - hy + (2 * h + l) * l;
|
|
952
|
+
return z * builtin_sqrt(ly + lx + hy + hx);
|
|
953
|
+
}
|
|
954
|
+
|
|
955
|
+
export function imul(x: f64, y: f64): f64 {
|
|
956
|
+
/*
|
|
957
|
+
* Wasm (MVP) and JS have different approaches for double->int conversions.
|
|
958
|
+
*
|
|
959
|
+
* For emulate JS conversion behavior and avoid trapping from wasm we should modulate by MAX_INT
|
|
960
|
+
* our float-point arguments before actual convertion to integers.
|
|
961
|
+
*/
|
|
962
|
+
if (!isFinite(x + y)) return 0;
|
|
963
|
+
return dtoi32(x) * dtoi32(y);
|
|
964
|
+
}
|
|
965
|
+
|
|
966
|
+
export function log(x: f64): f64 { // see: musl/src/math/log.c and SUN COPYRIGHT NOTICE above
|
|
967
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
968
|
+
return log_lut(x);
|
|
969
|
+
} else {
|
|
970
|
+
const
|
|
971
|
+
ln2_hi = reinterpret<f64>(0x3FE62E42FEE00000), // 6.93147180369123816490e-01
|
|
972
|
+
ln2_lo = reinterpret<f64>(0x3DEA39EF35793C76), // 1.90821492927058770002e-10
|
|
973
|
+
Lg1 = reinterpret<f64>(0x3FE5555555555593), // 6.666666666666735130e-01
|
|
974
|
+
Lg2 = reinterpret<f64>(0x3FD999999997FA04), // 3.999999999940941908e-01
|
|
975
|
+
Lg3 = reinterpret<f64>(0x3FD2492494229359), // 2.857142874366239149e-01
|
|
976
|
+
Lg4 = reinterpret<f64>(0x3FCC71C51D8E78AF), // 2.222219843214978396e-01
|
|
977
|
+
Lg5 = reinterpret<f64>(0x3FC7466496CB03DE), // 1.818357216161805012e-01
|
|
978
|
+
Lg6 = reinterpret<f64>(0x3FC39A09D078C69F), // 1.531383769920937332e-01
|
|
979
|
+
Lg7 = reinterpret<f64>(0x3FC2F112DF3E5244), // 1.479819860511658591e-01
|
|
980
|
+
Ox1p54 = reinterpret<f64>(0x4350000000000000); // 0x1p54
|
|
981
|
+
|
|
982
|
+
let u = reinterpret<u64>(x);
|
|
983
|
+
let hx = u32(u >> 32);
|
|
984
|
+
let k = 0;
|
|
985
|
+
let sign = hx >> 31;
|
|
986
|
+
if (sign || hx < 0x00100000) {
|
|
987
|
+
if (u << 1 == 0) return -1 / (x * x);
|
|
988
|
+
if (sign) return (x - x) / 0.0;
|
|
989
|
+
k -= 54;
|
|
990
|
+
x *= Ox1p54;
|
|
991
|
+
u = reinterpret<u64>(x);
|
|
992
|
+
hx = u32(u >> 32);
|
|
993
|
+
} else if (hx >= 0x7FF00000) {
|
|
994
|
+
return x;
|
|
995
|
+
} else if (hx == 0x3FF00000 && u << 32 == 0) {
|
|
996
|
+
return 0;
|
|
997
|
+
}
|
|
998
|
+
hx += 0x3FF00000 - 0x3FE6A09E;
|
|
999
|
+
k += (<i32>hx >> 20) - 0x3FF;
|
|
1000
|
+
hx = (hx & 0x000FFFFF) + 0x3FE6A09E;
|
|
1001
|
+
u = <u64>hx << 32 | (u & 0xFFFFFFFF);
|
|
1002
|
+
x = reinterpret<f64>(u);
|
|
1003
|
+
let f = x - 1.0;
|
|
1004
|
+
let hfsq = 0.5 * f * f;
|
|
1005
|
+
let s = f / (2.0 + f);
|
|
1006
|
+
let z = s * s;
|
|
1007
|
+
let w = z * z;
|
|
1008
|
+
let t1 = w * (Lg2 + w * (Lg4 + w * Lg6));
|
|
1009
|
+
let t2 = z * (Lg1 + w * (Lg3 + w * (Lg5 + w * Lg7)));
|
|
1010
|
+
let r = t2 + t1;
|
|
1011
|
+
let dk = <f64>k;
|
|
1012
|
+
return s * (hfsq + r) + dk * ln2_lo - hfsq + f + dk * ln2_hi;
|
|
1013
|
+
}
|
|
1014
|
+
}
|
|
1015
|
+
|
|
1016
|
+
export function log10(x: f64): f64 { // see: musl/src/math/log10.c and SUN COPYRIGHT NOTICE above
|
|
1017
|
+
const
|
|
1018
|
+
ivln10hi = reinterpret<f64>(0x3FDBCB7B15200000), // 4.34294481878168880939e-01
|
|
1019
|
+
ivln10lo = reinterpret<f64>(0x3DBB9438CA9AADD5), // 2.50829467116452752298e-11
|
|
1020
|
+
log10_2hi = reinterpret<f64>(0x3FD34413509F6000), // 3.01029995663611771306e-01
|
|
1021
|
+
log10_2lo = reinterpret<f64>(0x3D59FEF311F12B36), // 3.69423907715893078616e-13
|
|
1022
|
+
Lg1 = reinterpret<f64>(0x3FE5555555555593), // 6.666666666666735130e-01
|
|
1023
|
+
Lg2 = reinterpret<f64>(0x3FD999999997FA04), // 3.999999999940941908e-01
|
|
1024
|
+
Lg3 = reinterpret<f64>(0x3FD2492494229359), // 2.857142874366239149e-01
|
|
1025
|
+
Lg4 = reinterpret<f64>(0x3FCC71C51D8E78AF), // 2.222219843214978396e-01
|
|
1026
|
+
Lg5 = reinterpret<f64>(0x3FC7466496CB03DE), // 1.818357216161805012e-01
|
|
1027
|
+
Lg6 = reinterpret<f64>(0x3FC39A09D078C69F), // 1.531383769920937332e-01
|
|
1028
|
+
Lg7 = reinterpret<f64>(0x3FC2F112DF3E5244), // 1.479819860511658591e-01
|
|
1029
|
+
Ox1p54 = reinterpret<f64>(0x4350000000000000); // 0x1p54
|
|
1030
|
+
|
|
1031
|
+
let u = reinterpret<u64>(x);
|
|
1032
|
+
let hx = u32(u >> 32);
|
|
1033
|
+
let k = 0;
|
|
1034
|
+
let sign = hx >> 31;
|
|
1035
|
+
if (sign || hx < 0x00100000) {
|
|
1036
|
+
if (u << 1 == 0) return -1 / (x * x);
|
|
1037
|
+
if (sign) return (x - x) / 0.0;
|
|
1038
|
+
k -= 54;
|
|
1039
|
+
x *= Ox1p54;
|
|
1040
|
+
u = reinterpret<u64>(x);
|
|
1041
|
+
hx = u32(u >> 32);
|
|
1042
|
+
} else if (hx >= 0x7FF00000) {
|
|
1043
|
+
return x;
|
|
1044
|
+
} else if (hx == 0x3FF00000 && u << 32 == 0) {
|
|
1045
|
+
return 0;
|
|
1046
|
+
}
|
|
1047
|
+
hx += 0x3FF00000 - 0x3FE6A09E;
|
|
1048
|
+
k += i32(hx >> 20) - 0x3FF;
|
|
1049
|
+
hx = (hx & 0x000FFFFF) + 0x3FE6A09E;
|
|
1050
|
+
u = <u64>hx << 32 | (u & 0xFFFFFFFF);
|
|
1051
|
+
x = reinterpret<f64>(u);
|
|
1052
|
+
let f = x - 1.0;
|
|
1053
|
+
let hfsq = 0.5 * f * f;
|
|
1054
|
+
let s = f / (2.0 + f);
|
|
1055
|
+
let z = s * s;
|
|
1056
|
+
let w = z * z;
|
|
1057
|
+
let t1 = w * (Lg2 + w * (Lg4 + w * Lg6));
|
|
1058
|
+
let t2 = z * (Lg1 + w * (Lg3 + w * (Lg5 + w * Lg7)));
|
|
1059
|
+
let r = t2 + t1;
|
|
1060
|
+
let hi = f - hfsq;
|
|
1061
|
+
u = reinterpret<u64>(hi);
|
|
1062
|
+
u &= 0xFFFFFFFF00000000;
|
|
1063
|
+
hi = reinterpret<f64>(u);
|
|
1064
|
+
let lo = f - hi - hfsq + s * (hfsq + r);
|
|
1065
|
+
let val_hi = hi * ivln10hi;
|
|
1066
|
+
let dk = <f64>k;
|
|
1067
|
+
let y = dk * log10_2hi;
|
|
1068
|
+
let val_lo = dk * log10_2lo + (lo + hi) * ivln10lo + lo * ivln10hi;
|
|
1069
|
+
w = y + val_hi;
|
|
1070
|
+
val_lo += (y - w) + val_hi;
|
|
1071
|
+
return val_lo + w;
|
|
1072
|
+
}
|
|
1073
|
+
|
|
1074
|
+
export function log1p(x: f64): f64 { // see: musl/src/math/log1p.c and SUN COPYRIGHT NOTICE above
|
|
1075
|
+
const
|
|
1076
|
+
ln2_hi = reinterpret<f64>(0x3FE62E42FEE00000), // 6.93147180369123816490e-01
|
|
1077
|
+
ln2_lo = reinterpret<f64>(0x3DEA39EF35793C76), // 1.90821492927058770002e-10
|
|
1078
|
+
Lg1 = reinterpret<f64>(0x3FE5555555555593), // 6.666666666666735130e-01
|
|
1079
|
+
Lg2 = reinterpret<f64>(0x3FD999999997FA04), // 3.999999999940941908e-01
|
|
1080
|
+
Lg3 = reinterpret<f64>(0x3FD2492494229359), // 2.857142874366239149e-01
|
|
1081
|
+
Lg4 = reinterpret<f64>(0x3FCC71C51D8E78AF), // 2.222219843214978396e-01
|
|
1082
|
+
Lg5 = reinterpret<f64>(0x3FC7466496CB03DE), // 1.818357216161805012e-01
|
|
1083
|
+
Lg6 = reinterpret<f64>(0x3FC39A09D078C69F), // 1.531383769920937332e-01
|
|
1084
|
+
Lg7 = reinterpret<f64>(0x3FC2F112DF3E5244); // 1.479819860511658591e-01
|
|
1085
|
+
|
|
1086
|
+
let u = reinterpret<u64>(x);
|
|
1087
|
+
let hx = u32(u >> 32);
|
|
1088
|
+
let k = 1;
|
|
1089
|
+
let c = 0.0, f = 0.0;
|
|
1090
|
+
if (hx < 0x3FDA827A || bool(hx >> 31)) {
|
|
1091
|
+
if (hx >= 0xBFF00000) {
|
|
1092
|
+
if (x == -1) return x / 0.0;
|
|
1093
|
+
return (x - x) / 0.0;
|
|
1094
|
+
}
|
|
1095
|
+
if (hx << 1 < 0x3CA00000 << 1) return x;
|
|
1096
|
+
if (hx <= 0xBFD2BEC4) {
|
|
1097
|
+
k = 0;
|
|
1098
|
+
c = 0;
|
|
1099
|
+
f = x;
|
|
1100
|
+
}
|
|
1101
|
+
} else if (hx >= 0x7FF00000) return x;
|
|
1102
|
+
if (k) {
|
|
1103
|
+
u = reinterpret<u64>(1 + x);
|
|
1104
|
+
let hu = u32(u >> 32);
|
|
1105
|
+
hu += 0x3FF00000 - 0x3FE6A09E;
|
|
1106
|
+
k = i32(hu >> 20) - 0x3FF;
|
|
1107
|
+
if (k < 54) {
|
|
1108
|
+
let uf = reinterpret<f64>(u);
|
|
1109
|
+
c = k >= 2 ? 1 - (uf - x) : x - (uf - 1);
|
|
1110
|
+
c /= uf;
|
|
1111
|
+
} else c = 0;
|
|
1112
|
+
hu = (hu & 0x000FFFFF) + 0x3FE6A09E;
|
|
1113
|
+
u = <u64>hu << 32 | (u & 0xFFFFFFFF);
|
|
1114
|
+
f = reinterpret<f64>(u) - 1;
|
|
1115
|
+
}
|
|
1116
|
+
let hfsq = 0.5 * f * f;
|
|
1117
|
+
let s = f / (2.0 + f);
|
|
1118
|
+
let z = s * s;
|
|
1119
|
+
let w = z * z;
|
|
1120
|
+
let t1 = w * (Lg2 + w * (Lg4 + w * Lg6));
|
|
1121
|
+
let t2 = z * (Lg1 + w * (Lg3 + w * (Lg5 + w * Lg7)));
|
|
1122
|
+
let r = t2 + t1;
|
|
1123
|
+
let dk = <f64>k;
|
|
1124
|
+
return s * (hfsq + r) + (dk * ln2_lo + c) - hfsq + f + dk * ln2_hi;
|
|
1125
|
+
}
|
|
1126
|
+
|
|
1127
|
+
export function log2(x: f64): f64 { // see: musl/src/math/log2.c and SUN COPYRIGHT NOTICE above
|
|
1128
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
1129
|
+
return log2_lut(x);
|
|
1130
|
+
} else {
|
|
1131
|
+
const
|
|
1132
|
+
ivln2hi = reinterpret<f64>(0x3FF7154765200000), // 1.44269504072144627571e+00
|
|
1133
|
+
ivln2lo = reinterpret<f64>(0x3DE705FC2EEFA200), // 1.67517131648865118353e-10
|
|
1134
|
+
Lg1 = reinterpret<f64>(0x3FE5555555555593), // 6.666666666666735130e-01
|
|
1135
|
+
Lg2 = reinterpret<f64>(0x3FD999999997FA04), // 3.999999999940941908e-01
|
|
1136
|
+
Lg3 = reinterpret<f64>(0x3FD2492494229359), // 2.857142874366239149e-01
|
|
1137
|
+
Lg4 = reinterpret<f64>(0x3FCC71C51D8E78AF), // 2.222219843214978396e-01
|
|
1138
|
+
Lg5 = reinterpret<f64>(0x3FC7466496CB03DE), // 1.818357216161805012e-01
|
|
1139
|
+
Lg6 = reinterpret<f64>(0x3FC39A09D078C69F), // 1.531383769920937332e-01
|
|
1140
|
+
Lg7 = reinterpret<f64>(0x3FC2F112DF3E5244), // 1.479819860511658591e-01
|
|
1141
|
+
Ox1p54 = reinterpret<f64>(0x4350000000000000); // 1p54
|
|
1142
|
+
|
|
1143
|
+
let u = reinterpret<u64>(x);
|
|
1144
|
+
let hx = u32(u >> 32);
|
|
1145
|
+
let k = 0;
|
|
1146
|
+
let sign = hx >> 31;
|
|
1147
|
+
if (sign || hx < 0x00100000) {
|
|
1148
|
+
if (u << 1 == 0) return -1 / (x * x);
|
|
1149
|
+
if (sign) return (x - x) / 0.0;
|
|
1150
|
+
k -= 54;
|
|
1151
|
+
x *= Ox1p54;
|
|
1152
|
+
u = reinterpret<u64>(x);
|
|
1153
|
+
hx = u32(u >> 32);
|
|
1154
|
+
} else if (hx >= 0x7FF00000) {
|
|
1155
|
+
return x;
|
|
1156
|
+
} else if (hx == 0x3FF00000 && u << 32 == 0) {
|
|
1157
|
+
return 0;
|
|
1158
|
+
}
|
|
1159
|
+
hx += 0x3FF00000 - 0x3FE6A09E;
|
|
1160
|
+
k += i32(hx >> 20) - 0x3FF;
|
|
1161
|
+
hx = (hx & 0x000FFFFF) + 0x3FE6A09E;
|
|
1162
|
+
u = <u64>hx << 32 | (u & 0xFFFFFFFF);
|
|
1163
|
+
x = reinterpret<f64>(u);
|
|
1164
|
+
let f = x - 1.0;
|
|
1165
|
+
let hfsq = 0.5 * f * f;
|
|
1166
|
+
let s = f / (2.0 + f);
|
|
1167
|
+
let z = s * s;
|
|
1168
|
+
let w = z * z;
|
|
1169
|
+
let t1 = w * (Lg2 + w * (Lg4 + w * Lg6));
|
|
1170
|
+
let t2 = z * (Lg1 + w * (Lg3 + w * (Lg5 + w * Lg7)));
|
|
1171
|
+
let r = t2 + t1;
|
|
1172
|
+
let hi = f - hfsq;
|
|
1173
|
+
u = reinterpret<u64>(hi);
|
|
1174
|
+
u &= 0xFFFFFFFF00000000;
|
|
1175
|
+
hi = reinterpret<f64>(u);
|
|
1176
|
+
let lo = f - hi - hfsq + s * (hfsq + r);
|
|
1177
|
+
let val_hi = hi * ivln2hi;
|
|
1178
|
+
let val_lo = (lo + hi) * ivln2lo + lo * ivln2hi;
|
|
1179
|
+
let y = <f64>k;
|
|
1180
|
+
w = y + val_hi;
|
|
1181
|
+
val_lo += (y - w) + val_hi;
|
|
1182
|
+
val_hi = w;
|
|
1183
|
+
return val_lo + val_hi;
|
|
1184
|
+
}
|
|
1185
|
+
}
|
|
1186
|
+
|
|
1187
|
+
// @ts-ignore: decorator
|
|
1188
|
+
@inline
|
|
1189
|
+
export function max(value1: f64, value2: f64): f64 {
|
|
1190
|
+
return builtin_max<f64>(value1, value2);
|
|
1191
|
+
}
|
|
1192
|
+
|
|
1193
|
+
// @ts-ignore: decorator
|
|
1194
|
+
@inline
|
|
1195
|
+
export function min(value1: f64, value2: f64): f64 {
|
|
1196
|
+
return builtin_min<f64>(value1, value2);
|
|
1197
|
+
}
|
|
1198
|
+
|
|
1199
|
+
export function pow(x: f64, y: f64): f64 { // see: musl/src/math/pow.c and SUN COPYRIGHT NOTICE above
|
|
1200
|
+
// TODO: remove this fast pathes after introduced own mid-end IR with "stdlib call simplify" transforms
|
|
1201
|
+
if (builtin_abs<f64>(y) <= 2) {
|
|
1202
|
+
if (y == 2.0) return x * x;
|
|
1203
|
+
if (y == 0.5) {
|
|
1204
|
+
return select<f64>(
|
|
1205
|
+
builtin_abs<f64>(builtin_sqrt<f64>(x)),
|
|
1206
|
+
Infinity,
|
|
1207
|
+
x != -Infinity
|
|
1208
|
+
);
|
|
1209
|
+
}
|
|
1210
|
+
if (y == -1.0) return 1 / x;
|
|
1211
|
+
if (y == 1.0) return x;
|
|
1212
|
+
if (y == 0.0) return 1.0;
|
|
1213
|
+
}
|
|
1214
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
1215
|
+
return pow_lut(x, y);
|
|
1216
|
+
} else {
|
|
1217
|
+
const
|
|
1218
|
+
dp_h1 = reinterpret<f64>(0x3FE2B80340000000), // 5.84962487220764160156e-01
|
|
1219
|
+
dp_l1 = reinterpret<f64>(0x3E4CFDEB43CFD006), // 1.35003920212974897128e-08
|
|
1220
|
+
two53 = reinterpret<f64>(0x4340000000000000), // 9007199254740992.0
|
|
1221
|
+
huge = reinterpret<f64>(0x7E37E43C8800759C), // 1e+300
|
|
1222
|
+
tiny = reinterpret<f64>(0x01A56E1FC2F8F359), // 1e-300
|
|
1223
|
+
L1 = reinterpret<f64>(0x3FE3333333333303), // 5.99999999999994648725e-01
|
|
1224
|
+
L2 = reinterpret<f64>(0x3FDB6DB6DB6FABFF), // 4.28571428578550184252e-01
|
|
1225
|
+
L3 = reinterpret<f64>(0x3FD55555518F264D), // 3.33333329818377432918e-01
|
|
1226
|
+
L4 = reinterpret<f64>(0x3FD17460A91D4101), // 2.72728123808534006489e-01
|
|
1227
|
+
L5 = reinterpret<f64>(0x3FCD864A93C9DB65), // 2.30660745775561754067e-01
|
|
1228
|
+
L6 = reinterpret<f64>(0x3FCA7E284A454EEF), // 2.06975017800338417784e-01
|
|
1229
|
+
P1 = reinterpret<f64>(0x3FC555555555553E), // 1.66666666666666019037e-01
|
|
1230
|
+
P2 = reinterpret<f64>(0xBF66C16C16BEBD93), // -2.77777777770155933842e-03
|
|
1231
|
+
P3 = reinterpret<f64>(0x3F11566AAF25DE2C), // 6.61375632143793436117e-05
|
|
1232
|
+
P4 = reinterpret<f64>(0xBEBBBD41C5D26BF1), // -1.65339022054652515390e-06
|
|
1233
|
+
P5 = reinterpret<f64>(0x3E66376972BEA4D0), // 4.13813679705723846039e-08
|
|
1234
|
+
lg2 = reinterpret<f64>(0x3FE62E42FEFA39EF), // 6.93147180559945286227e-01
|
|
1235
|
+
lg2_h = reinterpret<f64>(0x3FE62E4300000000), // 6.93147182464599609375e-01
|
|
1236
|
+
lg2_l = reinterpret<f64>(0xBE205C610CA86C39), // -1.90465429995776804525e-09
|
|
1237
|
+
ovt = reinterpret<f64>(0x3C971547652B82FE), // 8.0085662595372944372e-017
|
|
1238
|
+
cp = reinterpret<f64>(0x3FEEC709DC3A03FD), // 9.61796693925975554329e-01
|
|
1239
|
+
cp_h = reinterpret<f64>(0x3FEEC709E0000000), // 9.61796700954437255859e-01
|
|
1240
|
+
cp_l = reinterpret<f64>(0xBE3E2FE0145B01F5), // -7.02846165095275826516e-09
|
|
1241
|
+
ivln2 = reinterpret<f64>(0x3FF71547652B82FE), // 1.44269504088896338700e+00
|
|
1242
|
+
ivln2_h = reinterpret<f64>(0x3FF7154760000000), // 1.44269502162933349609e+00
|
|
1243
|
+
ivln2_l = reinterpret<f64>(0x3E54AE0BF85DDF44), // 1.92596299112661746887e-08
|
|
1244
|
+
inv3 = reinterpret<f64>(0x3FD5555555555555); // 0.3333333333333333333333
|
|
1245
|
+
|
|
1246
|
+
let u_ = reinterpret<u64>(x);
|
|
1247
|
+
let hx = i32(u_ >> 32);
|
|
1248
|
+
let lx = <u32>u_;
|
|
1249
|
+
u_ = reinterpret<u64>(y);
|
|
1250
|
+
let hy = i32(u_ >> 32);
|
|
1251
|
+
let ly = <u32>u_;
|
|
1252
|
+
let ix = hx & 0x7FFFFFFF;
|
|
1253
|
+
let iy = hy & 0x7FFFFFFF;
|
|
1254
|
+
if ((iy | ly) == 0) return 1.0; // x**0 = 1, even if x is NaN
|
|
1255
|
+
// if (hx == 0x3FF00000 && lx == 0) return 1.0; // C: 1**y = 1, even if y is NaN, JS: NaN
|
|
1256
|
+
if ( // NaN if either arg is NaN
|
|
1257
|
+
ix > 0x7FF00000 || (ix == 0x7FF00000 && lx != 0) ||
|
|
1258
|
+
iy > 0x7FF00000 || (iy == 0x7FF00000 && ly != 0)
|
|
1259
|
+
) return x + y;
|
|
1260
|
+
let yisint = 0, k: i32;
|
|
1261
|
+
if (hx < 0) {
|
|
1262
|
+
if (iy >= 0x43400000) yisint = 2;
|
|
1263
|
+
else if (iy >= 0x3FF00000) {
|
|
1264
|
+
k = (iy >> 20) - 0x3FF;
|
|
1265
|
+
let offset = select<u32>(52, 20, k > 20) - k;
|
|
1266
|
+
let Ly = select<u32>(ly, iy, k > 20);
|
|
1267
|
+
let jj = Ly >> offset;
|
|
1268
|
+
if ((jj << offset) == Ly) yisint = 2 - (jj & 1);
|
|
1269
|
+
}
|
|
1270
|
+
}
|
|
1271
|
+
if (ly == 0) {
|
|
1272
|
+
if (iy == 0x7FF00000) { // y is +-inf
|
|
1273
|
+
if (((ix - 0x3FF00000) | lx) == 0) return NaN; // C: (-1)**+-inf is 1, JS: NaN
|
|
1274
|
+
else if (ix >= 0x3FF00000) return hy >= 0 ? y : 0.0; // (|x|>1)**+-inf = inf,0
|
|
1275
|
+
else return hy >= 0 ? 0.0 : -y; // (|x|<1)**+-inf = 0,inf
|
|
1276
|
+
}
|
|
1277
|
+
if (iy == 0x3FF00000) {
|
|
1278
|
+
if (hy >= 0) return x;
|
|
1279
|
+
return 1 / x;
|
|
1280
|
+
}
|
|
1281
|
+
if (hy == 0x40000000) return x * x;
|
|
1282
|
+
if (hy == 0x3FE00000) {
|
|
1283
|
+
if (hx >= 0) return builtin_sqrt(x);
|
|
1284
|
+
}
|
|
1285
|
+
}
|
|
1286
|
+
let ax = builtin_abs<f64>(x), z: f64;
|
|
1287
|
+
if (lx == 0) {
|
|
1288
|
+
if (ix == 0 || ix == 0x7FF00000 || ix == 0x3FF00000) {
|
|
1289
|
+
z = ax;
|
|
1290
|
+
if (hy < 0) z = 1.0 / z;
|
|
1291
|
+
if (hx < 0) {
|
|
1292
|
+
if (((ix - 0x3FF00000) | yisint) == 0) {
|
|
1293
|
+
let d = z - z;
|
|
1294
|
+
z = d / d;
|
|
1295
|
+
} else if (yisint == 1) z = -z;
|
|
1296
|
+
}
|
|
1297
|
+
return z;
|
|
1298
|
+
}
|
|
1299
|
+
}
|
|
1300
|
+
let s = 1.0;
|
|
1301
|
+
if (hx < 0) {
|
|
1302
|
+
if (yisint == 0) {
|
|
1303
|
+
let d = x - x;
|
|
1304
|
+
return d / d;
|
|
1305
|
+
}
|
|
1306
|
+
if (yisint == 1) s = -1.0;
|
|
1307
|
+
}
|
|
1308
|
+
let t1: f64, t2: f64, p_h: f64, p_l: f64, r: f64, t: f64, u: f64, v: f64, w: f64;
|
|
1309
|
+
let j: i32, n: i32;
|
|
1310
|
+
if (iy > 0x41E00000) {
|
|
1311
|
+
if (iy > 0x43F00000) {
|
|
1312
|
+
if (ix <= 0x3FEFFFFF) return hy < 0 ? huge * huge : tiny * tiny;
|
|
1313
|
+
if (ix >= 0x3FF00000) return hy > 0 ? huge * huge : tiny * tiny;
|
|
1314
|
+
}
|
|
1315
|
+
if (ix < 0x3FEFFFFF) return hy < 0 ? s * huge * huge : s * tiny * tiny;
|
|
1316
|
+
if (ix > 0x3FF00000) return hy > 0 ? s * huge * huge : s * tiny * tiny;
|
|
1317
|
+
t = ax - 1.0;
|
|
1318
|
+
w = (t * t) * (0.5 - t * (inv3 - t * 0.25));
|
|
1319
|
+
u = ivln2_h * t;
|
|
1320
|
+
v = t * ivln2_l - w * ivln2;
|
|
1321
|
+
t1 = u + v;
|
|
1322
|
+
t1 = reinterpret<f64>(reinterpret<u64>(t1) & 0xFFFFFFFF00000000);
|
|
1323
|
+
t2 = v - (t1 - u);
|
|
1324
|
+
} else {
|
|
1325
|
+
let ss: f64, s2: f64, s_h: f64, s_l: f64, t_h: f64, t_l: f64;
|
|
1326
|
+
n = 0;
|
|
1327
|
+
if (ix < 0x00100000) {
|
|
1328
|
+
ax *= two53;
|
|
1329
|
+
n -= 53;
|
|
1330
|
+
ix = <u32>(reinterpret<u64>(ax) >> 32);
|
|
1331
|
+
}
|
|
1332
|
+
n += (ix >> 20) - 0x3FF;
|
|
1333
|
+
j = ix & 0x000FFFFF;
|
|
1334
|
+
ix = j | 0x3FF00000;
|
|
1335
|
+
if (j <= 0x3988E) k = 0;
|
|
1336
|
+
else if (j < 0xBB67A) k = 1;
|
|
1337
|
+
else {
|
|
1338
|
+
k = 0;
|
|
1339
|
+
n += 1;
|
|
1340
|
+
ix -= 0x00100000;
|
|
1341
|
+
}
|
|
1342
|
+
ax = reinterpret<f64>(reinterpret<u64>(ax) & 0xFFFFFFFF | (<u64>ix << 32));
|
|
1343
|
+
let bp = select<f64>(1.5, 1.0, k); // k ? 1.5 : 1.0
|
|
1344
|
+
u = ax - bp;
|
|
1345
|
+
v = 1.0 / (ax + bp);
|
|
1346
|
+
ss = u * v;
|
|
1347
|
+
s_h = ss;
|
|
1348
|
+
s_h = reinterpret<f64>(reinterpret<u64>(s_h) & 0xFFFFFFFF00000000);
|
|
1349
|
+
t_h = reinterpret<f64>(u64(((ix >> 1) | 0x20000000) + 0x00080000 + (k << 18)) << 32);
|
|
1350
|
+
t_l = ax - (t_h - bp);
|
|
1351
|
+
s_l = v * ((u - s_h * t_h) - s_h * t_l);
|
|
1352
|
+
s2 = ss * ss;
|
|
1353
|
+
r = s2 * s2 * (L1 + s2 * (L2 + s2 * (L3 + s2 * (L4 + s2 * (L5 + s2 * L6)))));
|
|
1354
|
+
r += s_l * (s_h + ss);
|
|
1355
|
+
s2 = s_h * s_h;
|
|
1356
|
+
t_h = 3.0 + s2 + r;
|
|
1357
|
+
t_h = reinterpret<f64>(reinterpret<u64>(t_h) & 0xFFFFFFFF00000000);
|
|
1358
|
+
t_l = r - ((t_h - 3.0) - s2);
|
|
1359
|
+
u = s_h * t_h;
|
|
1360
|
+
v = s_l * t_h + t_l * ss;
|
|
1361
|
+
p_h = u + v;
|
|
1362
|
+
p_h = reinterpret<f64>(reinterpret<u64>(p_h) & 0xFFFFFFFF00000000);
|
|
1363
|
+
p_l = v - (p_h - u);
|
|
1364
|
+
let z_h = cp_h * p_h;
|
|
1365
|
+
let dp_l = select<f64>(dp_l1, 0.0, k);
|
|
1366
|
+
let z_l = cp_l * p_h + p_l * cp + dp_l;
|
|
1367
|
+
t = <f64>n;
|
|
1368
|
+
let dp_h = select<f64>(dp_h1, 0.0, k);
|
|
1369
|
+
t1 = ((z_h + z_l) + dp_h) + t;
|
|
1370
|
+
t1 = reinterpret<f64>(reinterpret<u64>(t1) & 0xFFFFFFFF00000000);
|
|
1371
|
+
t2 = z_l - (((t1 - t) - dp_h) - z_h);
|
|
1372
|
+
}
|
|
1373
|
+
let y1 = y;
|
|
1374
|
+
y1 = reinterpret<f64>(reinterpret<u64>(y1) & 0xFFFFFFFF00000000);
|
|
1375
|
+
p_l = (y - y1) * t1 + y * t2;
|
|
1376
|
+
p_h = y1 * t1;
|
|
1377
|
+
z = p_l + p_h;
|
|
1378
|
+
u_ = reinterpret<u64>(z);
|
|
1379
|
+
j = u32(u_ >> 32);
|
|
1380
|
+
let i = <i32>u_;
|
|
1381
|
+
if (j >= 0x40900000) {
|
|
1382
|
+
if (((j - 0x40900000) | i) != 0) return s * huge * huge;
|
|
1383
|
+
if (p_l + ovt > z - p_h) return s * huge * huge;
|
|
1384
|
+
} else if ((j & 0x7FFFFFFF) >= 0x4090CC00) {
|
|
1385
|
+
if (((j - 0xC090CC00) | i) != 0) return s * tiny * tiny;
|
|
1386
|
+
if (p_l <= z - p_h) return s * tiny * tiny;
|
|
1387
|
+
}
|
|
1388
|
+
i = j & 0x7FFFFFFF;
|
|
1389
|
+
k = (i >> 20) - 0x3FF;
|
|
1390
|
+
n = 0;
|
|
1391
|
+
if (i > 0x3FE00000) {
|
|
1392
|
+
n = j + (0x00100000 >> (k + 1));
|
|
1393
|
+
k = ((n & 0x7FFFFFFF) >> 20) - 0x3FF;
|
|
1394
|
+
t = 0.0;
|
|
1395
|
+
t = reinterpret<f64>(u64(n & ~(0x000FFFFF >> k)) << 32);
|
|
1396
|
+
n = ((n & 0x000FFFFF) | 0x00100000) >> (20 - k);
|
|
1397
|
+
if (j < 0) n = -n;
|
|
1398
|
+
p_h -= t;
|
|
1399
|
+
}
|
|
1400
|
+
t = p_l + p_h;
|
|
1401
|
+
t = reinterpret<f64>(reinterpret<u64>(t) & 0xFFFFFFFF00000000);
|
|
1402
|
+
u = t * lg2_h;
|
|
1403
|
+
v = (p_l - (t - p_h)) * lg2 + t * lg2_l;
|
|
1404
|
+
z = u + v;
|
|
1405
|
+
w = v - (z - u);
|
|
1406
|
+
t = z * z;
|
|
1407
|
+
t1 = z - t * (P1 + t * (P2 + t * (P3 + t * (P4 + t * P5))));
|
|
1408
|
+
r = (z * t1) / (t1 - 2.0) - (w + z * w);
|
|
1409
|
+
z = 1.0 - (r - z);
|
|
1410
|
+
j = u32(reinterpret<u64>(z) >> 32);
|
|
1411
|
+
j += n << 20;
|
|
1412
|
+
if ((j >> 20) <= 0) z = scalbn(z, n);
|
|
1413
|
+
else z = reinterpret<f64>(reinterpret<u64>(z) & 0xFFFFFFFF | (<u64>j << 32));
|
|
1414
|
+
return s * z;
|
|
1415
|
+
}
|
|
1416
|
+
}
|
|
1417
|
+
|
|
1418
|
+
export function seedRandom(value: i64): void {
|
|
1419
|
+
// Instead zero seed use golden ratio:
|
|
1420
|
+
// phi = (1 + sqrt(5)) / 2
|
|
1421
|
+
// trunc(2^64 / phi) = 0x9e3779b97f4a7c15
|
|
1422
|
+
if (value == 0) value = 0x9e3779b97f4a7c15;
|
|
1423
|
+
random_state0_64 = murmurHash3(value);
|
|
1424
|
+
random_state1_64 = murmurHash3(~random_state0_64);
|
|
1425
|
+
random_state0_32 = splitMix32(<u32>value);
|
|
1426
|
+
random_state1_32 = splitMix32(random_state0_32);
|
|
1427
|
+
random_seeded = true;
|
|
1428
|
+
}
|
|
1429
|
+
|
|
1430
|
+
export function random(): f64 { // see: v8/src/base/utils/random-number-generator.cc
|
|
1431
|
+
if (!random_seeded) seedRandom(reinterpret<i64>(seed()));
|
|
1432
|
+
let s1 = random_state0_64;
|
|
1433
|
+
let s0 = random_state1_64;
|
|
1434
|
+
random_state0_64 = s0;
|
|
1435
|
+
s1 ^= s1 << 23;
|
|
1436
|
+
s1 ^= s1 >> 17;
|
|
1437
|
+
s1 ^= s0;
|
|
1438
|
+
s1 ^= s0 >> 26;
|
|
1439
|
+
random_state1_64 = s1;
|
|
1440
|
+
let r = (s0 >> 12) | 0x3FF0000000000000;
|
|
1441
|
+
return reinterpret<f64>(r) - 1;
|
|
1442
|
+
}
|
|
1443
|
+
|
|
1444
|
+
export function round(x: f64): f64 {
|
|
1445
|
+
if (ASC_SHRINK_LEVEL > 0) {
|
|
1446
|
+
return builtin_ceil<f64>(x) - f64(builtin_ceil<f64>(x) - 0.5 > x);
|
|
1447
|
+
} else {
|
|
1448
|
+
let roundUp = builtin_ceil<f64>(x);
|
|
1449
|
+
return select<f64>(roundUp, roundUp - 1.0, roundUp - 0.5 <= x);
|
|
1450
|
+
}
|
|
1451
|
+
}
|
|
1452
|
+
|
|
1453
|
+
export function sign(x: f64): f64 {
|
|
1454
|
+
if (ASC_SHRINK_LEVEL > 0) {
|
|
1455
|
+
return select<f64>(builtin_copysign<f64>(1, x), x, builtin_abs(x) > 0);
|
|
1456
|
+
} else {
|
|
1457
|
+
return select<f64>(1, select<f64>(-1, x, x < 0), x > 0);
|
|
1458
|
+
}
|
|
1459
|
+
}
|
|
1460
|
+
|
|
1461
|
+
// @ts-ignore: decorator
|
|
1462
|
+
@inline
|
|
1463
|
+
export function signbit(x: f64): bool {
|
|
1464
|
+
return bool(reinterpret<u64>(x) >>> 63);
|
|
1465
|
+
}
|
|
1466
|
+
|
|
1467
|
+
export function sin(x: f64): f64 { // see: musl/src/math/sin.c
|
|
1468
|
+
let u = reinterpret<u64>(x);
|
|
1469
|
+
let ux = u32(u >> 32);
|
|
1470
|
+
let sign = ux >> 31;
|
|
1471
|
+
|
|
1472
|
+
ux &= 0x7FFFFFFF;
|
|
1473
|
+
|
|
1474
|
+
// |x| ~< pi/4
|
|
1475
|
+
if (ux <= 0x3FE921FB) {
|
|
1476
|
+
if (ux < 0x3E500000) { // |x| < 2**-26
|
|
1477
|
+
return x;
|
|
1478
|
+
}
|
|
1479
|
+
return sin_kern(x, 0.0, 0);
|
|
1480
|
+
}
|
|
1481
|
+
|
|
1482
|
+
// sin(Inf or NaN) is NaN
|
|
1483
|
+
if (ux >= 0x7FF00000) return x - x;
|
|
1484
|
+
|
|
1485
|
+
// argument reduction needed
|
|
1486
|
+
let n = rempio2(x, u, sign);
|
|
1487
|
+
let y0 = rempio2_y0;
|
|
1488
|
+
let y1 = rempio2_y1;
|
|
1489
|
+
|
|
1490
|
+
x = n & 1 ? cos_kern(y0, y1) : sin_kern(y0, y1, 1);
|
|
1491
|
+
return n & 2 ? -x : x;
|
|
1492
|
+
}
|
|
1493
|
+
|
|
1494
|
+
export function sinh(x: f64): f64 { // see: musl/src/math/sinh.c
|
|
1495
|
+
let u = reinterpret<u64>(x) & 0x7FFFFFFFFFFFFFFF;
|
|
1496
|
+
let a = reinterpret<f64>(u);
|
|
1497
|
+
let w = u32(u >> 32);
|
|
1498
|
+
let h = builtin_copysign(0.5, x);
|
|
1499
|
+
if (w < 0x40862E42) {
|
|
1500
|
+
let t = expm1(a);
|
|
1501
|
+
if (w < 0x3FF00000) {
|
|
1502
|
+
if (w < 0x3FF00000 - (26 << 20)) return x;
|
|
1503
|
+
return h * (2 * t - t * t / (t + 1));
|
|
1504
|
+
}
|
|
1505
|
+
return h * (t + t / (t + 1));
|
|
1506
|
+
}
|
|
1507
|
+
return expo2(a, 2 * h);
|
|
1508
|
+
}
|
|
1509
|
+
|
|
1510
|
+
// @ts-ignore: decorator
|
|
1511
|
+
@inline
|
|
1512
|
+
export function sqrt(x: f64): f64 {
|
|
1513
|
+
return builtin_sqrt<f64>(x);
|
|
1514
|
+
}
|
|
1515
|
+
|
|
1516
|
+
export function tan(x: f64): f64 { // see: musl/src/math/tan.c
|
|
1517
|
+
let u = reinterpret<u64>(x);
|
|
1518
|
+
let ux = u32(u >> 32);
|
|
1519
|
+
let sign = ux >>> 31;
|
|
1520
|
+
|
|
1521
|
+
ux &= 0x7FFFFFFF;
|
|
1522
|
+
|
|
1523
|
+
// |x| ~< pi/4
|
|
1524
|
+
if (ux <= 0x3FE921FB) {
|
|
1525
|
+
if (ux < 0x3E400000) { // |x| < 2**-27
|
|
1526
|
+
return x;
|
|
1527
|
+
}
|
|
1528
|
+
return tan_kern(x, 0.0, 1);
|
|
1529
|
+
}
|
|
1530
|
+
|
|
1531
|
+
// tan(Inf or NaN) is NaN
|
|
1532
|
+
if (ux >= 0x7FF00000) return x - x;
|
|
1533
|
+
|
|
1534
|
+
let n = rempio2(x, u, sign);
|
|
1535
|
+
return tan_kern(rempio2_y0, rempio2_y1, 1 - ((n & 1) << 1));
|
|
1536
|
+
}
|
|
1537
|
+
|
|
1538
|
+
export function tanh(x: f64): f64 { // see: musl/src/math/tanh.c
|
|
1539
|
+
let u = reinterpret<u64>(x);
|
|
1540
|
+
u &= 0x7FFFFFFFFFFFFFFF;
|
|
1541
|
+
let y = reinterpret<f64>(u);
|
|
1542
|
+
let w = u32(u >> 32);
|
|
1543
|
+
let t: f64;
|
|
1544
|
+
if (w > 0x3FE193EA) {
|
|
1545
|
+
if (w > 0x40340000) {
|
|
1546
|
+
t = 1 - 0 / y;
|
|
1547
|
+
} else {
|
|
1548
|
+
t = expm1(2 * y);
|
|
1549
|
+
t = 1 - 2 / (t + 2);
|
|
1550
|
+
}
|
|
1551
|
+
} else if (w > 0x3FD058AE) {
|
|
1552
|
+
t = expm1(2 * y);
|
|
1553
|
+
t = t / (t + 2);
|
|
1554
|
+
} else if (w >= 0x00100000) {
|
|
1555
|
+
t = expm1(-2 * y);
|
|
1556
|
+
t = -t / (t + 2);
|
|
1557
|
+
} else t = y;
|
|
1558
|
+
return builtin_copysign<f64>(t, x);
|
|
1559
|
+
}
|
|
1560
|
+
|
|
1561
|
+
// @ts-ignore: decorator
|
|
1562
|
+
@inline
|
|
1563
|
+
export function trunc(x: f64): f64 {
|
|
1564
|
+
return builtin_trunc<f64>(x);
|
|
1565
|
+
}
|
|
1566
|
+
|
|
1567
|
+
export function scalbn(x: f64, n: i32): f64 { // see: https://git.musl-libc.org/cgit/musl/tree/src/math/scalbn.c
|
|
1568
|
+
const
|
|
1569
|
+
Ox1p53 = reinterpret<f64>(0x4340000000000000),
|
|
1570
|
+
Ox1p1023 = reinterpret<f64>(0x7FE0000000000000),
|
|
1571
|
+
Ox1p_1022 = reinterpret<f64>(0x0010000000000000);
|
|
1572
|
+
|
|
1573
|
+
let y = x;
|
|
1574
|
+
if (n > 1023) {
|
|
1575
|
+
y *= Ox1p1023;
|
|
1576
|
+
n -= 1023;
|
|
1577
|
+
if (n > 1023) {
|
|
1578
|
+
y *= Ox1p1023;
|
|
1579
|
+
n = builtin_min<i32>(n - 1023, 1023);
|
|
1580
|
+
}
|
|
1581
|
+
} else if (n < -1022) {
|
|
1582
|
+
// make sure final n < -53 to avoid double
|
|
1583
|
+
// rounding in the subnormal range
|
|
1584
|
+
y *= Ox1p_1022 * Ox1p53;
|
|
1585
|
+
n += 1022 - 53;
|
|
1586
|
+
if (n < -1022) {
|
|
1587
|
+
y *= Ox1p_1022 * Ox1p53;
|
|
1588
|
+
n = builtin_max<i32>(n + 1022 - 53, -1022);
|
|
1589
|
+
}
|
|
1590
|
+
}
|
|
1591
|
+
return y * reinterpret<f64>(<u64>(0x3FF + n) << 52);
|
|
1592
|
+
}
|
|
1593
|
+
|
|
1594
|
+
export function mod(x: f64, y: f64): f64 { // see: musl/src/math/fmod.c
|
|
1595
|
+
if (builtin_abs<f64>(y) == 1.0) {
|
|
1596
|
+
// x % 1, x % -1 ==> sign(x) * abs(x - 1.0 * trunc(x / 1.0))
|
|
1597
|
+
// TODO: move this rule to compiler's optimization pass.
|
|
1598
|
+
// It could be apply for any x % C_pot, where "C_pot" is pow of two const.
|
|
1599
|
+
return builtin_copysign<f64>(x - builtin_trunc<f64>(x), x);
|
|
1600
|
+
}
|
|
1601
|
+
let ux = reinterpret<u64>(x);
|
|
1602
|
+
let uy = reinterpret<u64>(y);
|
|
1603
|
+
let ex = i64(ux >> 52 & 0x7FF);
|
|
1604
|
+
let ey = i64(uy >> 52 & 0x7FF);
|
|
1605
|
+
let sx = ux >> 63;
|
|
1606
|
+
let uy1 = uy << 1;
|
|
1607
|
+
if (uy1 == 0 || ex == 0x7FF || isNaN<f64>(y)) {
|
|
1608
|
+
let m = x * y;
|
|
1609
|
+
return m / m;
|
|
1610
|
+
}
|
|
1611
|
+
let ux1 = ux << 1;
|
|
1612
|
+
if (ux1 <= uy1) {
|
|
1613
|
+
return x * f64(ux1 != uy1);
|
|
1614
|
+
}
|
|
1615
|
+
if (!ex) {
|
|
1616
|
+
ex -= builtin_clz<i64>(ux << 12);
|
|
1617
|
+
ux <<= 1 - ex;
|
|
1618
|
+
} else {
|
|
1619
|
+
ux &= u64(-1) >> 12;
|
|
1620
|
+
ux |= 1 << 52;
|
|
1621
|
+
}
|
|
1622
|
+
if (!ey) {
|
|
1623
|
+
ey -= builtin_clz<i64>(uy << 12);
|
|
1624
|
+
uy <<= 1 - ey;
|
|
1625
|
+
} else {
|
|
1626
|
+
uy &= u64(-1) >> 12;
|
|
1627
|
+
uy |= 1 << 52;
|
|
1628
|
+
}
|
|
1629
|
+
while (ex > ey) {
|
|
1630
|
+
if (ux >= uy) {
|
|
1631
|
+
if (ux == uy) return 0 * x;
|
|
1632
|
+
ux -= uy;
|
|
1633
|
+
}
|
|
1634
|
+
ux <<= 1;
|
|
1635
|
+
--ex;
|
|
1636
|
+
}
|
|
1637
|
+
if (ux >= uy) {
|
|
1638
|
+
if (ux == uy) return 0 * x;
|
|
1639
|
+
ux -= uy;
|
|
1640
|
+
}
|
|
1641
|
+
// for (; !(ux >> 52); ux <<= 1) --ex;
|
|
1642
|
+
let shift = builtin_clz<i64>(ux << 11);
|
|
1643
|
+
ex -= shift;
|
|
1644
|
+
ux <<= shift;
|
|
1645
|
+
if (ex > 0) {
|
|
1646
|
+
ux -= 1 << 52;
|
|
1647
|
+
ux |= ex << 52;
|
|
1648
|
+
} else {
|
|
1649
|
+
ux >>= -ex + 1;
|
|
1650
|
+
}
|
|
1651
|
+
return reinterpret<f64>(ux | (sx << 63));
|
|
1652
|
+
}
|
|
1653
|
+
|
|
1654
|
+
export function rem(x: f64, y: f64): f64 { // see: musl/src/math/remquo.c
|
|
1655
|
+
let ux = reinterpret<u64>(x);
|
|
1656
|
+
let uy = reinterpret<u64>(y);
|
|
1657
|
+
let ex = i64(ux >> 52 & 0x7FF);
|
|
1658
|
+
let ey = i64(uy >> 52 & 0x7FF);
|
|
1659
|
+
if (uy << 1 == 0 || ex == 0x7FF || isNaN(y)) {
|
|
1660
|
+
let m = x * y;
|
|
1661
|
+
return m / m;
|
|
1662
|
+
}
|
|
1663
|
+
if (ux << 1 == 0) return x;
|
|
1664
|
+
let uxi = ux;
|
|
1665
|
+
if (!ex) {
|
|
1666
|
+
ex -= builtin_clz<i64>(uxi << 12);
|
|
1667
|
+
uxi <<= 1 - ex;
|
|
1668
|
+
} else {
|
|
1669
|
+
uxi &= u64(-1) >> 12;
|
|
1670
|
+
uxi |= 1 << 52;
|
|
1671
|
+
}
|
|
1672
|
+
if (!ey) {
|
|
1673
|
+
ey -= builtin_clz<i64>(uy << 12);
|
|
1674
|
+
uy <<= 1 - ey;
|
|
1675
|
+
} else {
|
|
1676
|
+
uy &= u64(-1) >> 12;
|
|
1677
|
+
uy |= 1 << 52;
|
|
1678
|
+
}
|
|
1679
|
+
let q: u32 = 0;
|
|
1680
|
+
do {
|
|
1681
|
+
if (ex < ey) {
|
|
1682
|
+
if (ex + 1 == ey) break; // goto end
|
|
1683
|
+
return x;
|
|
1684
|
+
}
|
|
1685
|
+
while (ex > ey) {
|
|
1686
|
+
if (uxi >= uy) {
|
|
1687
|
+
uxi -= uy;
|
|
1688
|
+
++q;
|
|
1689
|
+
}
|
|
1690
|
+
uxi <<= 1;
|
|
1691
|
+
q <<= 1;
|
|
1692
|
+
--ex;
|
|
1693
|
+
}
|
|
1694
|
+
if (uxi >= uy) {
|
|
1695
|
+
uxi -= uy;
|
|
1696
|
+
++q;
|
|
1697
|
+
}
|
|
1698
|
+
if (uxi == 0) ex = -60;
|
|
1699
|
+
else {
|
|
1700
|
+
let shift = builtin_clz<i64>(uxi << 11);
|
|
1701
|
+
ex -= shift;
|
|
1702
|
+
uxi <<= shift;
|
|
1703
|
+
}
|
|
1704
|
+
break;
|
|
1705
|
+
} while (false);
|
|
1706
|
+
// end:
|
|
1707
|
+
if (ex > 0) {
|
|
1708
|
+
uxi -= 1 << 52;
|
|
1709
|
+
uxi |= ex << 52;
|
|
1710
|
+
} else {
|
|
1711
|
+
uxi >>= -ex + 1;
|
|
1712
|
+
}
|
|
1713
|
+
x = reinterpret<f64>(uxi);
|
|
1714
|
+
y = builtin_abs<f64>(y);
|
|
1715
|
+
let x2 = x + x;
|
|
1716
|
+
if (ex == ey || (ex + 1 == ey && (x2 > y || (x2 == y && <bool>(q & 1))))) {
|
|
1717
|
+
x -= y;
|
|
1718
|
+
// ++q;
|
|
1719
|
+
}
|
|
1720
|
+
return <i64>ux < 0 ? -x : x;
|
|
1721
|
+
}
|
|
1722
|
+
|
|
1723
|
+
export function sincos(x: f64): void { // see: musl/tree/src/math/sincos.c
|
|
1724
|
+
let u = reinterpret<u64>(x);
|
|
1725
|
+
let ux = u32(u >> 32);
|
|
1726
|
+
let sign = ux >> 31;
|
|
1727
|
+
ux &= 0x7FFFFFFF;
|
|
1728
|
+
|
|
1729
|
+
if (ux <= 0x3FE921FB) { // |x| ~<= π/4
|
|
1730
|
+
if (ux < 0x3E46A09E) { // if |x| < 2**-27 * sqrt(2)
|
|
1731
|
+
sincos_sin = x;
|
|
1732
|
+
sincos_cos = 1;
|
|
1733
|
+
return;
|
|
1734
|
+
}
|
|
1735
|
+
sincos_sin = sin_kern(x, 0, 0);
|
|
1736
|
+
sincos_cos = cos_kern(x, 0);
|
|
1737
|
+
return;
|
|
1738
|
+
}
|
|
1739
|
+
// sin(Inf or NaN) is NaN
|
|
1740
|
+
if (ux >= 0x7F800000) {
|
|
1741
|
+
let xx = x - x;
|
|
1742
|
+
sincos_sin = xx;
|
|
1743
|
+
sincos_cos = xx;
|
|
1744
|
+
return;
|
|
1745
|
+
}
|
|
1746
|
+
// general argument reduction needed
|
|
1747
|
+
let n = rempio2(x, u, sign);
|
|
1748
|
+
let y0 = rempio2_y0;
|
|
1749
|
+
let y1 = rempio2_y1;
|
|
1750
|
+
let s = sin_kern(y0, y1, 1);
|
|
1751
|
+
let c = cos_kern(y0, y1);
|
|
1752
|
+
let sin = s, cos = c;
|
|
1753
|
+
if (n & 1) {
|
|
1754
|
+
sin = c;
|
|
1755
|
+
cos = -s;
|
|
1756
|
+
}
|
|
1757
|
+
if (n & 2) {
|
|
1758
|
+
sin = -sin;
|
|
1759
|
+
cos = -cos;
|
|
1760
|
+
}
|
|
1761
|
+
sincos_sin = sin;
|
|
1762
|
+
sincos_cos = cos;
|
|
1763
|
+
}
|
|
1764
|
+
}
|
|
1765
|
+
|
|
1766
|
+
// @ts-ignore: decorator
|
|
1767
|
+
@lazy let rempio2f_y: f64;
|
|
1768
|
+
|
|
1769
|
+
// @ts-ignore: decorator
|
|
1770
|
+
@lazy @inline const PIO2F_TABLE = memory.data<u64>([
|
|
1771
|
+
0xA2F9836E4E441529,
|
|
1772
|
+
0xFC2757D1F534DDC0,
|
|
1773
|
+
0xDB6295993C439041,
|
|
1774
|
+
0xFE5163ABDEBBC561
|
|
1775
|
+
]);
|
|
1776
|
+
|
|
1777
|
+
function Rf(z: f32): f32 { // Rational approximation of (asin(x)-x)/x^3
|
|
1778
|
+
const // see: musl/src/math/asinf.c and SUN COPYRIGHT NOTICE above
|
|
1779
|
+
pS0 = reinterpret<f32>(0x3E2AAA75), // 1.6666586697e-01f
|
|
1780
|
+
pS1 = reinterpret<f32>(0xBD2F13BA), // -4.2743422091e-02f
|
|
1781
|
+
pS2 = reinterpret<f32>(0xBC0DD36B), // -8.6563630030e-03f
|
|
1782
|
+
qS1 = reinterpret<f32>(0xBF34E5AE); // -7.0662963390e-01f
|
|
1783
|
+
|
|
1784
|
+
let p = z * (pS0 + z * (pS1 + z * pS2));
|
|
1785
|
+
let q: f32 = 1 + z * qS1;
|
|
1786
|
+
return p / q;
|
|
1787
|
+
}
|
|
1788
|
+
|
|
1789
|
+
// @ts-ignore: decorator
|
|
1790
|
+
@inline
|
|
1791
|
+
function expo2f(x: f32, sign: f32): f32 { // exp(x)/2 for x >= log(DBL_MAX)
|
|
1792
|
+
const // see: musl/src/math/__expo2f.c
|
|
1793
|
+
k = <u32>235,
|
|
1794
|
+
kln2 = reinterpret<f32>(0x4322E3BC); // 0x1.45c778p+7f
|
|
1795
|
+
let scale = reinterpret<f32>(u32(0x7F + (k >> 1)) << 23);
|
|
1796
|
+
// in directed rounding correct sign before rounding or overflow is important
|
|
1797
|
+
return NativeMathf.exp(x - kln2) * (sign * scale) * scale;
|
|
1798
|
+
}
|
|
1799
|
+
|
|
1800
|
+
// @ts-ignore: decorator
|
|
1801
|
+
@inline
|
|
1802
|
+
function pio2f_large_quot(x: f32, u: i32): i32 { // see: jdh8/metallic/blob/master/src/math/float/rem_pio2f.c
|
|
1803
|
+
const coeff = reinterpret<f64>(0x3BF921FB54442D18); // π * 0x1p-65 = 8.51530395021638647334e-20
|
|
1804
|
+
|
|
1805
|
+
let offset = (u >> 23) - 152;
|
|
1806
|
+
let shift = u64(offset & 63);
|
|
1807
|
+
let tblPtr = PIO2F_TABLE + (offset >> 6 << 3);
|
|
1808
|
+
|
|
1809
|
+
let b0 = load<u64>(tblPtr, 0 << 3);
|
|
1810
|
+
let b1 = load<u64>(tblPtr, 1 << 3);
|
|
1811
|
+
let lo: u64;
|
|
1812
|
+
|
|
1813
|
+
if (shift > 32) {
|
|
1814
|
+
let b2 = load<u64>(tblPtr, 2 << 3);
|
|
1815
|
+
lo = b2 >> (96 - shift);
|
|
1816
|
+
lo |= b1 << (shift - 32);
|
|
1817
|
+
} else {
|
|
1818
|
+
lo = b1 >> (32 - shift);
|
|
1819
|
+
}
|
|
1820
|
+
|
|
1821
|
+
let hi = (b1 >> (64 - shift)) | (b0 << shift);
|
|
1822
|
+
let mantissa: u64 = (u & 0x007FFFFF) | 0x00800000;
|
|
1823
|
+
let product = mantissa * hi + (mantissa * lo >> 32);
|
|
1824
|
+
let r: i64 = product << 2;
|
|
1825
|
+
let q = i32((product >> 62) + (r >>> 63));
|
|
1826
|
+
rempio2f_y = copysign<f64>(coeff, x) * <f64>r;
|
|
1827
|
+
return q;
|
|
1828
|
+
}
|
|
1829
|
+
|
|
1830
|
+
// @ts-ignore: decorator
|
|
1831
|
+
@inline
|
|
1832
|
+
function rempio2f(x: f32, u: u32, sign: i32): i32 { // see: jdh8/metallic/blob/master/src/math/float/rem_pio2f.c
|
|
1833
|
+
const
|
|
1834
|
+
pi2hi = reinterpret<f64>(0x3FF921FB50000000), // 1.57079631090164184570
|
|
1835
|
+
pi2lo = reinterpret<f64>(0x3E5110B4611A6263), // 1.58932547735281966916e-8
|
|
1836
|
+
_2_pi = reinterpret<f64>(0x3FE45F306DC9C883); // 0.63661977236758134308
|
|
1837
|
+
|
|
1838
|
+
if (u < 0x4DC90FDB) { // π * 0x1p28
|
|
1839
|
+
let q = nearest(x * _2_pi);
|
|
1840
|
+
rempio2f_y = x - q * pi2hi - q * pi2lo;
|
|
1841
|
+
return <i32>q;
|
|
1842
|
+
}
|
|
1843
|
+
|
|
1844
|
+
let q = pio2f_large_quot(x, u);
|
|
1845
|
+
return select(-q, q, sign);
|
|
1846
|
+
}
|
|
1847
|
+
|
|
1848
|
+
// |sin(x)/x - s(x)| < 2**-37.5 (~[-4.89e-12, 4.824e-12]).
|
|
1849
|
+
// @ts-ignore: decorator
|
|
1850
|
+
@inline
|
|
1851
|
+
function sin_kernf(x: f64): f32 { // see: musl/tree/src/math/__sindf.c
|
|
1852
|
+
const
|
|
1853
|
+
S1 = reinterpret<f64>(0xBFC5555554CBAC77), // -0x15555554cbac77.0p-55
|
|
1854
|
+
S2 = reinterpret<f64>(0x3F811110896EFBB2), // 0x111110896efbb2.0p-59
|
|
1855
|
+
S3 = reinterpret<f64>(0xBF2A00F9E2CAE774), // -0x1a00f9e2cae774.0p-65
|
|
1856
|
+
S4 = reinterpret<f64>(0x3EC6CD878C3B46A7); // 0x16cd878c3b46a7.0p-71
|
|
1857
|
+
|
|
1858
|
+
let z = x * x;
|
|
1859
|
+
let w = z * z;
|
|
1860
|
+
let r = S3 + z * S4;
|
|
1861
|
+
let s = z * x;
|
|
1862
|
+
return f32((x + s * (S1 + z * S2)) + s * w * r);
|
|
1863
|
+
}
|
|
1864
|
+
|
|
1865
|
+
// |cos(x) - c(x)| < 2**-34.1 (~[-5.37e-11, 5.295e-11]).
|
|
1866
|
+
// @ts-ignore: decorator
|
|
1867
|
+
@inline
|
|
1868
|
+
function cos_kernf(x: f64): f32 { // see: musl/tree/src/math/__cosdf.c
|
|
1869
|
+
const
|
|
1870
|
+
C0 = reinterpret<f64>(0xBFDFFFFFFD0C5E81), // -0x1ffffffd0c5e81.0p-54
|
|
1871
|
+
C1 = reinterpret<f64>(0x3FA55553E1053A42), // 0x155553e1053a42.0p-57
|
|
1872
|
+
C2 = reinterpret<f64>(0xBF56C087E80F1E27), // -0x16c087e80f1e27.0p-62
|
|
1873
|
+
C3 = reinterpret<f64>(0x3EF99342E0EE5069); // 0x199342e0ee5069.0p-68
|
|
1874
|
+
|
|
1875
|
+
let z = x * x;
|
|
1876
|
+
let w = z * z;
|
|
1877
|
+
let r = C2 + z * C3;
|
|
1878
|
+
return f32(((1 + z * C0) + w * C1) + (w * z) * r);
|
|
1879
|
+
}
|
|
1880
|
+
|
|
1881
|
+
// |tan(x)/x - t(x)| < 2**-25.5 (~[-2e-08, 2e-08]).
|
|
1882
|
+
// @ts-ignore: decorator
|
|
1883
|
+
@inline
|
|
1884
|
+
function tan_kernf(x: f64, odd: i32): f32 { // see: musl/tree/src/math/__tandf.c
|
|
1885
|
+
const
|
|
1886
|
+
T0 = reinterpret<f64>(0x3FD5554D3418C99F), // 0x15554d3418c99f.0p-54
|
|
1887
|
+
T1 = reinterpret<f64>(0x3FC112FD38999F72), // 0x1112fd38999f72.0p-55
|
|
1888
|
+
T2 = reinterpret<f64>(0x3FAB54C91D865AFE), // 0x1b54c91d865afe.0p-57
|
|
1889
|
+
T3 = reinterpret<f64>(0x3F991DF3908C33CE), // 0x191df3908c33ce.0p-58
|
|
1890
|
+
T4 = reinterpret<f64>(0x3F685DADFCECF44E), // 0x185dadfcecf44e.0p-61
|
|
1891
|
+
T5 = reinterpret<f64>(0x3F8362B9BF971BCD); // 0x1362b9bf971bcd.0p-59
|
|
1892
|
+
|
|
1893
|
+
let z = x * x;
|
|
1894
|
+
let r = T4 + z * T5;
|
|
1895
|
+
let t = T2 + z * T3;
|
|
1896
|
+
let w = z * z;
|
|
1897
|
+
let s = z * x;
|
|
1898
|
+
let u = T0 + z * T1;
|
|
1899
|
+
|
|
1900
|
+
r = (x + s * u) + (s * w) * (t + w * r);
|
|
1901
|
+
return f32(odd ? -1 / r : r);
|
|
1902
|
+
}
|
|
1903
|
+
|
|
1904
|
+
// See: jdh8/metallic/src/math/float/log2f.c and jdh8/metallic/src/math/float/kernel/atanh.h
|
|
1905
|
+
// @ts-ignore: decorator
|
|
1906
|
+
@inline
|
|
1907
|
+
function log2f(x: f64): f64 {
|
|
1908
|
+
const
|
|
1909
|
+
log2e = reinterpret<f64>(0x3FF71547652B82FE), // 1.44269504088896340736
|
|
1910
|
+
c0 = reinterpret<f64>(0x3FD555554FD9CAEF), // 0.33333332822728226129
|
|
1911
|
+
c1 = reinterpret<f64>(0x3FC999A7A8AF4132), // 0.20000167595436263505
|
|
1912
|
+
c2 = reinterpret<f64>(0x3FC2438D79437030), // 0.14268654271188685375
|
|
1913
|
+
c3 = reinterpret<f64>(0x3FBE2F663B001C97); // 0.11791075649681414150
|
|
1914
|
+
|
|
1915
|
+
let i = reinterpret<i64>(x);
|
|
1916
|
+
let exponent = (i - 0x3FE6A09E667F3BCD) >> 52;
|
|
1917
|
+
x = reinterpret<f64>(i - (exponent << 52));
|
|
1918
|
+
x = (x - 1) / (x + 1);
|
|
1919
|
+
let xx = x * x;
|
|
1920
|
+
let y = x + x * xx * (c0 + c1 * xx + (c2 + c3 * xx) * (xx * xx));
|
|
1921
|
+
return (2 * log2e) * y + <f64>exponent;
|
|
1922
|
+
}
|
|
1923
|
+
|
|
1924
|
+
// See: jdh8/metallic/src/math/float/exp2f.h and jdh8/metallic/blob/master/src/math/float/kernel/exp2f.h
|
|
1925
|
+
// @ts-ignore: decorator
|
|
1926
|
+
@inline
|
|
1927
|
+
function exp2f(x: f64): f64 {
|
|
1928
|
+
const
|
|
1929
|
+
c0 = reinterpret<f64>(0x3FE62E4302FCC24A), // 6.931471880289532425e-1
|
|
1930
|
+
c1 = reinterpret<f64>(0x3FCEBFBE07D97B91), // 2.402265108421173406e-1
|
|
1931
|
+
c2 = reinterpret<f64>(0x3FAC6AF6CCFC1A65), // 5.550357105498874537e-2
|
|
1932
|
+
c3 = reinterpret<f64>(0x3F83B29E3CE9AEF6), // 9.618030771171497658e-3
|
|
1933
|
+
c4 = reinterpret<f64>(0x3F55F0896145A89F), // 1.339086685300950937e-3
|
|
1934
|
+
c5 = reinterpret<f64>(0x3F2446C81E384864); // 1.546973499989028719e-4
|
|
1935
|
+
|
|
1936
|
+
if (x < -1022) return 0;
|
|
1937
|
+
if (x >= 1024) return Infinity;
|
|
1938
|
+
|
|
1939
|
+
let n = nearest(x);
|
|
1940
|
+
x -= n;
|
|
1941
|
+
let xx = x * x;
|
|
1942
|
+
let y = 1 + x * (c0 + c1 * x + (c2 + c3 * x) * xx + (c4 + c5 * x) * (xx * xx));
|
|
1943
|
+
return reinterpret<f64>(reinterpret<i64>(y) + (<i64>n << 52));
|
|
1944
|
+
}
|
|
1945
|
+
|
|
1946
|
+
export namespace NativeMathf {
|
|
1947
|
+
|
|
1948
|
+
// @ts-ignore: decorator
|
|
1949
|
+
@lazy
|
|
1950
|
+
export const E = <f32>NativeMath.E;
|
|
1951
|
+
|
|
1952
|
+
// @ts-ignore: decorator
|
|
1953
|
+
@lazy
|
|
1954
|
+
export const LN2 = <f32>NativeMath.LN2;
|
|
1955
|
+
|
|
1956
|
+
// @ts-ignore: decorator
|
|
1957
|
+
@lazy
|
|
1958
|
+
export const LN10 = <f32>NativeMath.LN10;
|
|
1959
|
+
|
|
1960
|
+
// @ts-ignore: decorator
|
|
1961
|
+
@lazy
|
|
1962
|
+
export const LOG2E = <f32>NativeMath.LOG2E;
|
|
1963
|
+
|
|
1964
|
+
// @ts-ignore: decorator
|
|
1965
|
+
@lazy
|
|
1966
|
+
export const LOG10E = <f32>NativeMath.LOG10E;
|
|
1967
|
+
|
|
1968
|
+
// @ts-ignore: decorator
|
|
1969
|
+
@lazy
|
|
1970
|
+
export const PI = <f32>NativeMath.PI;
|
|
1971
|
+
|
|
1972
|
+
// @ts-ignore: decorator
|
|
1973
|
+
@lazy
|
|
1974
|
+
export const SQRT1_2 = <f32>NativeMath.SQRT1_2;
|
|
1975
|
+
|
|
1976
|
+
// @ts-ignore: decorator
|
|
1977
|
+
@lazy
|
|
1978
|
+
export const SQRT2 = <f32>NativeMath.SQRT2;
|
|
1979
|
+
|
|
1980
|
+
// @ts-ignore: decorator
|
|
1981
|
+
@lazy
|
|
1982
|
+
export let sincos_sin: f32 = 0;
|
|
1983
|
+
|
|
1984
|
+
// @ts-ignore: decorator
|
|
1985
|
+
@lazy
|
|
1986
|
+
export let sincos_cos: f32 = 0;
|
|
1987
|
+
|
|
1988
|
+
// @ts-ignore: decorator
|
|
1989
|
+
@inline
|
|
1990
|
+
export function abs(x: f32): f32 {
|
|
1991
|
+
return builtin_abs<f32>(x);
|
|
1992
|
+
}
|
|
1993
|
+
|
|
1994
|
+
export function acos(x: f32): f32 { // see: musl/src/math/acosf.c and SUN COPYRIGHT NOTICE above
|
|
1995
|
+
const
|
|
1996
|
+
pio2_hi = reinterpret<f32>(0x3FC90FDA), // 1.5707962513e+00f
|
|
1997
|
+
pio2_lo = reinterpret<f32>(0x33A22168), // 7.5497894159e-08f
|
|
1998
|
+
Ox1p_120f = reinterpret<f32>(0x03800000); // 0x1p-120f
|
|
1999
|
+
|
|
2000
|
+
let hx = reinterpret<u32>(x);
|
|
2001
|
+
let ix = hx & 0x7FFFFFFF;
|
|
2002
|
+
if (ix >= 0x3F800000) {
|
|
2003
|
+
if (ix == 0x3F800000) {
|
|
2004
|
+
return select<f32>(2 * pio2_hi + Ox1p_120f, 0, <i32>hx < 0);
|
|
2005
|
+
}
|
|
2006
|
+
return 0 / (x - x);
|
|
2007
|
+
}
|
|
2008
|
+
if (ix < 0x3F000000) {
|
|
2009
|
+
if (ix <= 0x32800000) return pio2_hi + Ox1p_120f;
|
|
2010
|
+
return pio2_hi - (x - (pio2_lo - x * Rf(x * x)));
|
|
2011
|
+
}
|
|
2012
|
+
let z: f32, w: f32, s: f32;
|
|
2013
|
+
if (<i32>hx < 0) {
|
|
2014
|
+
// z = (1 + x) * 0.5;
|
|
2015
|
+
z = 0.5 + x * 0.5;
|
|
2016
|
+
s = builtin_sqrt<f32>(z);
|
|
2017
|
+
w = Rf(z) * s - pio2_lo;
|
|
2018
|
+
return 2 * (pio2_hi - (s + w));
|
|
2019
|
+
}
|
|
2020
|
+
// z = (1 - x) * 0.5;
|
|
2021
|
+
z = 0.5 - x * 0.5;
|
|
2022
|
+
s = builtin_sqrt<f32>(z);
|
|
2023
|
+
hx = reinterpret<u32>(s);
|
|
2024
|
+
let df = reinterpret<f32>(hx & 0xFFFFF000);
|
|
2025
|
+
let c = (z - df * df) / (s + df);
|
|
2026
|
+
w = Rf(z) * s + c;
|
|
2027
|
+
return 2 * (df + w);
|
|
2028
|
+
}
|
|
2029
|
+
|
|
2030
|
+
export function acosh(x: f32): f32 { // see: musl/src/math/acoshf.c
|
|
2031
|
+
const s = reinterpret<f32>(0x3F317218); // 0.693147180559945309417232121458176568f
|
|
2032
|
+
let u = reinterpret<u32>(x);
|
|
2033
|
+
let a = u & 0x7FFFFFFF;
|
|
2034
|
+
if (a < 0x3F800000 + (1 << 23)) { // |x| < 2, invalid if x < 1
|
|
2035
|
+
let xm1 = x - 1;
|
|
2036
|
+
return log1p(xm1 + builtin_sqrt(xm1 * (xm1 + 2)));
|
|
2037
|
+
}
|
|
2038
|
+
if (u < 0x3F800000 + (12 << 23)) { // 2 <= x < 0x1p12
|
|
2039
|
+
return log(2 * x - 1 / (x + builtin_sqrt<f32>(x * x - 1)));
|
|
2040
|
+
}
|
|
2041
|
+
// x >= 0x1p12 or x <= -2 or NaN
|
|
2042
|
+
return log(x) + s;
|
|
2043
|
+
}
|
|
2044
|
+
|
|
2045
|
+
export function asin(x: f32): f32 { // see: musl/src/math/asinf.c and SUN COPYRIGHT NOTICE above
|
|
2046
|
+
const
|
|
2047
|
+
pio2 = reinterpret<f32>(0x3FC90FDB), // 1.570796326794896558e+00f
|
|
2048
|
+
Ox1p_120f = reinterpret<f32>(0x03800000); // 0x1p-120f
|
|
2049
|
+
|
|
2050
|
+
let sx = x;
|
|
2051
|
+
let hx = reinterpret<u32>(x) & 0x7FFFFFFF;
|
|
2052
|
+
if (hx >= 0x3F800000) {
|
|
2053
|
+
if (hx == 0x3F800000) return x * pio2 + Ox1p_120f;
|
|
2054
|
+
return 0 / (x - x);
|
|
2055
|
+
}
|
|
2056
|
+
if (hx < 0x3F000000) {
|
|
2057
|
+
if (hx < 0x39800000 && hx >= 0x00800000) return x;
|
|
2058
|
+
return x + x * Rf(x * x);
|
|
2059
|
+
}
|
|
2060
|
+
// let z: f32 = (1 - builtin_abs<f32>(x)) * 0.5;
|
|
2061
|
+
let z: f32 = 0.5 - builtin_abs<f32>(x) * 0.5;
|
|
2062
|
+
let s = builtin_sqrt<f64>(z); // sic
|
|
2063
|
+
x = f32(pio2 - 2 * (s + s * Rf(z)));
|
|
2064
|
+
return builtin_copysign(x, sx);
|
|
2065
|
+
}
|
|
2066
|
+
|
|
2067
|
+
export function asinh(x: f32): f32 { // see: musl/src/math/asinhf.c
|
|
2068
|
+
const c = reinterpret<f32>(0x3F317218); // 0.693147180559945309417232121458176568f
|
|
2069
|
+
let u = reinterpret<u32>(x) & 0x7FFFFFFF;
|
|
2070
|
+
let y = reinterpret<f32>(u);
|
|
2071
|
+
if (u >= 0x3F800000 + (12 << 23)) y = log(y) + c;
|
|
2072
|
+
else if (u >= 0x3F800000 + (1 << 23)) y = log(2 * y + 1 / (builtin_sqrt<f32>(y * y + 1) + y));
|
|
2073
|
+
else if (u >= 0x3F800000 - (12 << 23)) y = log1p(y + y * y / (builtin_sqrt<f32>(y * y + 1) + 1));
|
|
2074
|
+
return builtin_copysign(y, x);
|
|
2075
|
+
}
|
|
2076
|
+
|
|
2077
|
+
export function atan(x: f32): f32 { // see: musl/src/math/atanf.c and SUN COPYRIGHT NOTICE above
|
|
2078
|
+
const
|
|
2079
|
+
atanhi0 = reinterpret<f32>(0x3EED6338), // 4.6364760399e-01f
|
|
2080
|
+
atanhi1 = reinterpret<f32>(0x3F490FDA), // 7.8539812565e-01f
|
|
2081
|
+
atanhi2 = reinterpret<f32>(0x3F7B985E), // 9.8279368877e-01f
|
|
2082
|
+
atanhi3 = reinterpret<f32>(0x3FC90FDA), // 1.5707962513e+00f
|
|
2083
|
+
atanlo0 = reinterpret<f32>(0x31AC3769), // 5.0121582440e-09f
|
|
2084
|
+
atanlo1 = reinterpret<f32>(0x33222168), // 3.7748947079e-08f
|
|
2085
|
+
atanlo2 = reinterpret<f32>(0x33140FB4), // 3.4473217170e-08f
|
|
2086
|
+
atanlo3 = reinterpret<f32>(0x33A22168), // 7.5497894159e-08f
|
|
2087
|
+
aT0 = reinterpret<f32>(0x3EAAAAA9), // 3.3333328366e-01f
|
|
2088
|
+
aT1 = reinterpret<f32>(0xBE4CCA98), // -1.9999158382e-01f
|
|
2089
|
+
aT2 = reinterpret<f32>(0x3E11F50D), // 1.4253635705e-01f
|
|
2090
|
+
aT3 = reinterpret<f32>(0xBDDA1247), // -1.0648017377e-01f
|
|
2091
|
+
aT4 = reinterpret<f32>(0x3D7CAC25), // 6.1687607318e-02f
|
|
2092
|
+
Ox1p_120f = reinterpret<f32>(0x03800000); // 0x1p-120f
|
|
2093
|
+
|
|
2094
|
+
let ix = reinterpret<u32>(x);
|
|
2095
|
+
let sx = x;
|
|
2096
|
+
ix &= 0x7FFFFFFF;
|
|
2097
|
+
let z: f32;
|
|
2098
|
+
if (ix >= 0x4C800000) {
|
|
2099
|
+
if (isNaN(x)) return x;
|
|
2100
|
+
z = atanhi3 + Ox1p_120f;
|
|
2101
|
+
return builtin_copysign(z, sx);
|
|
2102
|
+
}
|
|
2103
|
+
let id: i32;
|
|
2104
|
+
if (ix < 0x3EE00000) {
|
|
2105
|
+
if (ix < 0x39800000) return x;
|
|
2106
|
+
id = -1;
|
|
2107
|
+
} else {
|
|
2108
|
+
x = builtin_abs<f32>(x);
|
|
2109
|
+
if (ix < 0x3F980000) {
|
|
2110
|
+
if (ix < 0x3F300000) {
|
|
2111
|
+
id = 0;
|
|
2112
|
+
x = (2.0 * x - 1.0) / (2.0 + x);
|
|
2113
|
+
} else {
|
|
2114
|
+
id = 1;
|
|
2115
|
+
x = (x - 1.0) / (x + 1.0);
|
|
2116
|
+
}
|
|
2117
|
+
} else {
|
|
2118
|
+
if (ix < 0x401C0000) {
|
|
2119
|
+
id = 2;
|
|
2120
|
+
x = (x - 1.5) / (1.0 + 1.5 * x);
|
|
2121
|
+
} else {
|
|
2122
|
+
id = 3;
|
|
2123
|
+
x = -1.0 / x;
|
|
2124
|
+
}
|
|
2125
|
+
}
|
|
2126
|
+
}
|
|
2127
|
+
z = x * x;
|
|
2128
|
+
let w = z * z;
|
|
2129
|
+
let s1 = z * (aT0 + w * (aT2 + w * aT4));
|
|
2130
|
+
let s2 = w * (aT1 + w * aT3);
|
|
2131
|
+
let s3 = x * (s1 + s2);
|
|
2132
|
+
if (id < 0) return x - s3;
|
|
2133
|
+
switch (id) {
|
|
2134
|
+
case 0: { z = atanhi0 - ((s3 - atanlo0) - x); break; }
|
|
2135
|
+
case 1: { z = atanhi1 - ((s3 - atanlo1) - x); break; }
|
|
2136
|
+
case 2: { z = atanhi2 - ((s3 - atanlo2) - x); break; }
|
|
2137
|
+
case 3: { z = atanhi3 - ((s3 - atanlo3) - x); break; }
|
|
2138
|
+
default: unreachable();
|
|
2139
|
+
}
|
|
2140
|
+
return builtin_copysign(z, sx);
|
|
2141
|
+
}
|
|
2142
|
+
|
|
2143
|
+
export function atanh(x: f32): f32 { // see: musl/src/math/atanhf.c
|
|
2144
|
+
let u = reinterpret<u32>(x);
|
|
2145
|
+
let y = builtin_abs(x);
|
|
2146
|
+
if (u < 0x3F800000 - (1 << 23)) {
|
|
2147
|
+
if (u >= 0x3F800000 - (32 << 23)) y = 0.5 * log1p(2 * y * (1.0 + y / (1 - y)));
|
|
2148
|
+
} else y = 0.5 * log1p(2 * (y / (1 - y)));
|
|
2149
|
+
return builtin_copysign(y, x);
|
|
2150
|
+
}
|
|
2151
|
+
|
|
2152
|
+
export function atan2(y: f32, x: f32): f32 { // see: musl/src/math/atan2f.c and SUN COPYRIGHT NOTICE above
|
|
2153
|
+
const
|
|
2154
|
+
pi = reinterpret<f32>(0x40490FDB), // 3.1415927410e+00f
|
|
2155
|
+
pi_lo = reinterpret<f32>(0xB3BBBD2E); // -8.7422776573e-08f
|
|
2156
|
+
|
|
2157
|
+
if (isNaN(x) || isNaN(y)) return x + y;
|
|
2158
|
+
let ix = reinterpret<u32>(x);
|
|
2159
|
+
let iy = reinterpret<u32>(y);
|
|
2160
|
+
if (ix == 0x3F800000) return atan(y);
|
|
2161
|
+
let m = u32(((iy >> 31) & 1) | ((ix >> 30) & 2));
|
|
2162
|
+
ix &= 0x7FFFFFFF;
|
|
2163
|
+
iy &= 0x7FFFFFFF;
|
|
2164
|
+
if (iy == 0) {
|
|
2165
|
+
switch (m) {
|
|
2166
|
+
case 0:
|
|
2167
|
+
case 1: return y;
|
|
2168
|
+
case 2: return pi;
|
|
2169
|
+
case 3: return -pi;
|
|
2170
|
+
}
|
|
2171
|
+
}
|
|
2172
|
+
if (ix == 0) return m & 1 ? -pi / 2 : pi / 2;
|
|
2173
|
+
if (ix == 0x7F800000) {
|
|
2174
|
+
if (iy == 0x7F800000) {
|
|
2175
|
+
let t: f32 = m & 2 ? 3 * pi / 4 : pi / 4;
|
|
2176
|
+
return m & 1 ? -t : t;
|
|
2177
|
+
} else {
|
|
2178
|
+
let t: f32 = m & 2 ? pi : 0.0;
|
|
2179
|
+
return m & 1 ? -t : t;
|
|
2180
|
+
}
|
|
2181
|
+
}
|
|
2182
|
+
if (ix + (26 << 23) < iy || iy == 0x7F800000) return m & 1 ? -pi / 2 : pi / 2;
|
|
2183
|
+
let z: f32;
|
|
2184
|
+
if ((m & 2) && iy + (26 << 23) < ix) z = 0.0;
|
|
2185
|
+
else z = atan(builtin_abs<f32>(y / x));
|
|
2186
|
+
switch (m) {
|
|
2187
|
+
case 0: return z;
|
|
2188
|
+
case 1: return -z;
|
|
2189
|
+
case 2: return pi - (z - pi_lo);
|
|
2190
|
+
case 3: return (z - pi_lo) - pi;
|
|
2191
|
+
}
|
|
2192
|
+
unreachable();
|
|
2193
|
+
return 0;
|
|
2194
|
+
}
|
|
2195
|
+
|
|
2196
|
+
export function cbrt(x: f32): f32 { // see: musl/src/math/cbrtf.c and SUN COPYRIGHT NOTICE above
|
|
2197
|
+
const
|
|
2198
|
+
B1 = <u32>709958130,
|
|
2199
|
+
B2 = <u32>642849266,
|
|
2200
|
+
Ox1p24f = reinterpret<f32>(0x4B800000);
|
|
2201
|
+
|
|
2202
|
+
let u = reinterpret<u32>(x);
|
|
2203
|
+
let hx = u & 0x7FFFFFFF;
|
|
2204
|
+
if (hx >= 0x7F800000) return x + x;
|
|
2205
|
+
if (hx < 0x00800000) {
|
|
2206
|
+
if (hx == 0) return x;
|
|
2207
|
+
u = reinterpret<u32>(x * Ox1p24f);
|
|
2208
|
+
hx = u & 0x7FFFFFFF;
|
|
2209
|
+
hx = hx / 3 + B2;
|
|
2210
|
+
} else {
|
|
2211
|
+
hx = hx / 3 + B1;
|
|
2212
|
+
}
|
|
2213
|
+
u &= 0x80000000;
|
|
2214
|
+
u |= hx;
|
|
2215
|
+
let t = <f64>reinterpret<f32>(u);
|
|
2216
|
+
let r = t * t * t;
|
|
2217
|
+
t = t * (<f64>x + x + r) / (x + r + r);
|
|
2218
|
+
r = t * t * t;
|
|
2219
|
+
t = t * (<f64>x + x + r) / (x + r + r);
|
|
2220
|
+
return <f32>t;
|
|
2221
|
+
}
|
|
2222
|
+
|
|
2223
|
+
// @ts-ignore: decorator
|
|
2224
|
+
@inline
|
|
2225
|
+
export function ceil(x: f32): f32 {
|
|
2226
|
+
return builtin_ceil<f32>(x);
|
|
2227
|
+
}
|
|
2228
|
+
|
|
2229
|
+
export function clz32(x: f32): f32 {
|
|
2230
|
+
if (!isFinite(x)) return 32;
|
|
2231
|
+
return <f32>builtin_clz(dtoi32(x));
|
|
2232
|
+
}
|
|
2233
|
+
|
|
2234
|
+
export function cos(x: f32): f32 { // see: musl/src/math/cosf.c
|
|
2235
|
+
const
|
|
2236
|
+
c1pio2 = reinterpret<f64>(0x3FF921FB54442D18), // M_PI_2 * 1
|
|
2237
|
+
c2pio2 = reinterpret<f64>(0x400921FB54442D18), // M_PI_2 * 2
|
|
2238
|
+
c3pio2 = reinterpret<f64>(0x4012D97C7F3321D2), // M_PI_2 * 3
|
|
2239
|
+
c4pio2 = reinterpret<f64>(0x401921FB54442D18); // M_PI_2 * 4
|
|
2240
|
+
|
|
2241
|
+
let ux = reinterpret<u32>(x);
|
|
2242
|
+
let sign = ux >> 31;
|
|
2243
|
+
ux &= 0x7FFFFFFF;
|
|
2244
|
+
|
|
2245
|
+
if (ux <= 0x3F490FDA) { // |x| ~<= π/4
|
|
2246
|
+
if (ux < 0x39800000) { // |x| < 2**-12
|
|
2247
|
+
// raise inexact if x != 0
|
|
2248
|
+
return 1;
|
|
2249
|
+
}
|
|
2250
|
+
return cos_kernf(x);
|
|
2251
|
+
}
|
|
2252
|
+
|
|
2253
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2254
|
+
if (ux <= 0x407B53D1) { // |x| ~<= 5π/4
|
|
2255
|
+
if (ux > 0x4016CBE3) { // |x| ~> 3π/4
|
|
2256
|
+
return -cos_kernf(sign ? x + c2pio2 : x - c2pio2);
|
|
2257
|
+
} else {
|
|
2258
|
+
return sign ? sin_kernf(x + c1pio2) : sin_kernf(c1pio2 - x);
|
|
2259
|
+
}
|
|
2260
|
+
}
|
|
2261
|
+
if (ux <= 0x40E231D5) { // |x| ~<= 9π/4
|
|
2262
|
+
if (ux > 0x40AFEDDF) { // |x| ~> 7π/4
|
|
2263
|
+
return cos_kernf(sign ? x + c4pio2 : x - c4pio2);
|
|
2264
|
+
} else {
|
|
2265
|
+
return sign ? sin_kernf(-x - c3pio2) : sin_kernf(x - c3pio2);
|
|
2266
|
+
}
|
|
2267
|
+
}
|
|
2268
|
+
}
|
|
2269
|
+
|
|
2270
|
+
// cos(Inf or NaN) is NaN
|
|
2271
|
+
if (ux >= 0x7F800000) return x - x;
|
|
2272
|
+
|
|
2273
|
+
// general argument reduction needed
|
|
2274
|
+
let n = rempio2f(x, ux, sign);
|
|
2275
|
+
let y = rempio2f_y;
|
|
2276
|
+
|
|
2277
|
+
let t = n & 1 ? sin_kernf(y) : cos_kernf(y);
|
|
2278
|
+
return (n + 1) & 2 ? -t : t;
|
|
2279
|
+
}
|
|
2280
|
+
|
|
2281
|
+
export function cosh(x: f32): f32 { // see: musl/src/math/coshf.c
|
|
2282
|
+
let u = reinterpret<u32>(x);
|
|
2283
|
+
u &= 0x7FFFFFFF;
|
|
2284
|
+
x = reinterpret<f32>(u);
|
|
2285
|
+
if (u < 0x3F317217) {
|
|
2286
|
+
if (u < 0x3F800000 - (12 << 23)) return 1;
|
|
2287
|
+
let t = expm1(x);
|
|
2288
|
+
// return 1 + t * t / (2 * (1 + t));
|
|
2289
|
+
return 1 + t * t / (2 + 2 * t);
|
|
2290
|
+
}
|
|
2291
|
+
if (u < 0x42B17217) {
|
|
2292
|
+
let t = exp(x);
|
|
2293
|
+
// return 0.5 * (t + 1 / t);
|
|
2294
|
+
return 0.5 * t + 0.5 / t;
|
|
2295
|
+
}
|
|
2296
|
+
return expo2f(x, 1);
|
|
2297
|
+
}
|
|
2298
|
+
|
|
2299
|
+
// @ts-ignore: decorator
|
|
2300
|
+
@inline
|
|
2301
|
+
export function floor(x: f32): f32 {
|
|
2302
|
+
return builtin_floor<f32>(x);
|
|
2303
|
+
}
|
|
2304
|
+
|
|
2305
|
+
export function exp(x: f32): f32 { // see: musl/src/math/expf.c and SUN COPYRIGHT NOTICE above
|
|
2306
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2307
|
+
return expf_lut(x);
|
|
2308
|
+
} else {
|
|
2309
|
+
const
|
|
2310
|
+
ln2hi = reinterpret<f32>(0x3F317200), // 6.9314575195e-1f
|
|
2311
|
+
ln2lo = reinterpret<f32>(0x35BFBE8E), // 1.4286067653e-6f
|
|
2312
|
+
invln2 = reinterpret<f32>(0x3FB8AA3B), // 1.4426950216e+0f
|
|
2313
|
+
P1 = reinterpret<f32>(0x3E2AAA8F), // 1.6666625440e-1f
|
|
2314
|
+
P2 = reinterpret<f32>(0xBB355215), // -2.7667332906e-3f
|
|
2315
|
+
Ox1p127f = reinterpret<f32>(0x7F000000); // 0x1p+127f
|
|
2316
|
+
|
|
2317
|
+
let hx = reinterpret<u32>(x);
|
|
2318
|
+
let sign = hx >> 31;
|
|
2319
|
+
hx &= 0x7FFFFFFF;
|
|
2320
|
+
if (hx >= 0x42AEAC50) {
|
|
2321
|
+
if (hx > 0x7F800000) return x; // NaN
|
|
2322
|
+
if (hx >= 0x42B17218) {
|
|
2323
|
+
if (!sign) return x * Ox1p127f;
|
|
2324
|
+
else if (hx >= 0x42CFF1B5) return 0;
|
|
2325
|
+
}
|
|
2326
|
+
}
|
|
2327
|
+
let hi: f32, lo: f32;
|
|
2328
|
+
let k: i32;
|
|
2329
|
+
if (hx > 0x3EB17218) {
|
|
2330
|
+
if (hx > 0x3F851592) {
|
|
2331
|
+
k = i32(invln2 * x + builtin_copysign<f32>(0.5, x));
|
|
2332
|
+
} else {
|
|
2333
|
+
k = 1 - (sign << 1);
|
|
2334
|
+
}
|
|
2335
|
+
hi = x - <f32>k * ln2hi;
|
|
2336
|
+
lo = <f32>k * ln2lo;
|
|
2337
|
+
x = hi - lo;
|
|
2338
|
+
} else if (hx > 0x39000000) {
|
|
2339
|
+
k = 0;
|
|
2340
|
+
hi = x;
|
|
2341
|
+
lo = 0;
|
|
2342
|
+
} else {
|
|
2343
|
+
return 1 + x;
|
|
2344
|
+
}
|
|
2345
|
+
let xx = x * x;
|
|
2346
|
+
let c = x - xx * (P1 + xx * P2);
|
|
2347
|
+
let y: f32 = 1 + (x * c / (2 - c) - lo + hi);
|
|
2348
|
+
return k == 0 ? y : scalbn(y, k);
|
|
2349
|
+
}
|
|
2350
|
+
}
|
|
2351
|
+
|
|
2352
|
+
export function exp2(x: f32): f32 {
|
|
2353
|
+
return exp2f_lut(x);
|
|
2354
|
+
}
|
|
2355
|
+
|
|
2356
|
+
export function expm1(x: f32): f32 { // see: musl/src/math/expm1f.c and SUN COPYRIGHT NOTICE above
|
|
2357
|
+
const
|
|
2358
|
+
ln2_hi = reinterpret<f32>(0x3F317180), // 6.9313812256e-01f
|
|
2359
|
+
ln2_lo = reinterpret<f32>(0x3717F7D1), // 9.0580006145e-06f
|
|
2360
|
+
invln2 = reinterpret<f32>(0x3FB8AA3B), // 1.4426950216e+00f
|
|
2361
|
+
Q1 = reinterpret<f32>(0xBD088868), // -3.3333212137e-02f
|
|
2362
|
+
Q2 = reinterpret<f32>(0x3ACF3010), // 1.5807170421e-03f
|
|
2363
|
+
Ox1p127f = reinterpret<f32>(0x7F000000); // 0x1p+127f
|
|
2364
|
+
|
|
2365
|
+
let u = reinterpret<u32>(x);
|
|
2366
|
+
let hx = u & 0x7FFFFFFF;
|
|
2367
|
+
let sign = u >> 31;
|
|
2368
|
+
if (hx >= 0x4195B844) {
|
|
2369
|
+
if (hx > 0x7F800000) return x;
|
|
2370
|
+
if (sign) return -1;
|
|
2371
|
+
if (hx > 0x42B17217) { // x > log(FLT_MAX)
|
|
2372
|
+
x *= Ox1p127f;
|
|
2373
|
+
return x;
|
|
2374
|
+
}
|
|
2375
|
+
}
|
|
2376
|
+
let c: f32 = 0.0, t: f32, k: i32;
|
|
2377
|
+
if (hx > 0x3EB17218) {
|
|
2378
|
+
k = select<i32>(
|
|
2379
|
+
1 - (sign << 1),
|
|
2380
|
+
i32(invln2 * x + builtin_copysign<f32>(0.5, x)),
|
|
2381
|
+
hx < 0x3F851592
|
|
2382
|
+
);
|
|
2383
|
+
t = <f32>k;
|
|
2384
|
+
let hi = x - t * ln2_hi;
|
|
2385
|
+
let lo = t * ln2_lo;
|
|
2386
|
+
x = hi - lo;
|
|
2387
|
+
c = (hi - x) - lo;
|
|
2388
|
+
} else if (hx < 0x33000000) {
|
|
2389
|
+
return x;
|
|
2390
|
+
} else k = 0;
|
|
2391
|
+
let hfx: f32 = 0.5 * x;
|
|
2392
|
+
let hxs: f32 = x * hfx;
|
|
2393
|
+
let r1: f32 = 1.0 + hxs * (Q1 + hxs * Q2);
|
|
2394
|
+
t = 3.0 - r1 * hfx;
|
|
2395
|
+
let e = hxs * ((r1 - t) / (6.0 - x * t));
|
|
2396
|
+
if (k == 0) return x - (x * e - hxs);
|
|
2397
|
+
e = x * (e - c) - c;
|
|
2398
|
+
e -= hxs;
|
|
2399
|
+
if (k == -1) return 0.5 * (x - e) - 0.5;
|
|
2400
|
+
if (k == 1) {
|
|
2401
|
+
if (x < -0.25) return -2.0 * (e - (x + 0.5));
|
|
2402
|
+
return 1.0 + 2.0 * (x - e);
|
|
2403
|
+
}
|
|
2404
|
+
u = (0x7F + k) << 23;
|
|
2405
|
+
let twopk = reinterpret<f32>(u);
|
|
2406
|
+
let y: f32;
|
|
2407
|
+
if (k < 0 || k > 56) {
|
|
2408
|
+
y = x - e + 1.0;
|
|
2409
|
+
if (k == 128) y = y * 2.0 * Ox1p127f;
|
|
2410
|
+
else y = y * twopk;
|
|
2411
|
+
return y - 1.0;
|
|
2412
|
+
}
|
|
2413
|
+
u = (0x7F - k) << 23;
|
|
2414
|
+
y = reinterpret<f32>(u);
|
|
2415
|
+
if (k < 20) y = (1 - y) - e;
|
|
2416
|
+
else y = 1 - (e + y);
|
|
2417
|
+
return (x + y) * twopk;
|
|
2418
|
+
}
|
|
2419
|
+
|
|
2420
|
+
// @ts-ignore: decorator
|
|
2421
|
+
@inline
|
|
2422
|
+
export function fround(x: f32): f32 {
|
|
2423
|
+
return x;
|
|
2424
|
+
}
|
|
2425
|
+
|
|
2426
|
+
export function hypot(x: f32, y: f32): f32 { // see: musl/src/math/hypotf.c
|
|
2427
|
+
const
|
|
2428
|
+
Ox1p90f = reinterpret<f32>(0x6C800000),
|
|
2429
|
+
Ox1p_90f = reinterpret<f32>(0x12800000);
|
|
2430
|
+
|
|
2431
|
+
let ux = reinterpret<u32>(x);
|
|
2432
|
+
let uy = reinterpret<u32>(y);
|
|
2433
|
+
ux &= 0x7FFFFFFF;
|
|
2434
|
+
uy &= 0x7FFFFFFF;
|
|
2435
|
+
if (ux < uy) {
|
|
2436
|
+
let ut = ux;
|
|
2437
|
+
ux = uy;
|
|
2438
|
+
uy = ut;
|
|
2439
|
+
}
|
|
2440
|
+
x = reinterpret<f32>(ux);
|
|
2441
|
+
y = reinterpret<f32>(uy);
|
|
2442
|
+
if (uy == 0xFF << 23) return y;
|
|
2443
|
+
if (ux >= 0xFF << 23 || uy == 0 || ux - uy >= 25 << 23) return x + y;
|
|
2444
|
+
let z: f32 = 1;
|
|
2445
|
+
if (ux >= (0x7F + 60) << 23) {
|
|
2446
|
+
z = Ox1p90f;
|
|
2447
|
+
x *= Ox1p_90f;
|
|
2448
|
+
y *= Ox1p_90f;
|
|
2449
|
+
} else if (uy < (0x7F - 60) << 23) {
|
|
2450
|
+
z = Ox1p_90f;
|
|
2451
|
+
x *= Ox1p90f;
|
|
2452
|
+
y *= Ox1p90f;
|
|
2453
|
+
}
|
|
2454
|
+
return z * builtin_sqrt<f32>(f32(<f64>x * x + <f64>y * y));
|
|
2455
|
+
}
|
|
2456
|
+
|
|
2457
|
+
// @ts-ignore: decorator
|
|
2458
|
+
@inline
|
|
2459
|
+
export function imul(x: f32, y: f32): f32 {
|
|
2460
|
+
/*
|
|
2461
|
+
* Wasm (MVP) and JS have different approaches for double->int conversions.
|
|
2462
|
+
*
|
|
2463
|
+
* For emulate JS conversion behavior and avoid trapping from wasm we should modulate by MAX_INT
|
|
2464
|
+
* our float-point arguments before actual convertion to integers.
|
|
2465
|
+
*/
|
|
2466
|
+
if (!isFinite(x + y)) return 0;
|
|
2467
|
+
return <f32>(dtoi32(x) * dtoi32(y));
|
|
2468
|
+
}
|
|
2469
|
+
|
|
2470
|
+
export function log(x: f32): f32 { // see: musl/src/math/logf.c and SUN COPYRIGHT NOTICE above
|
|
2471
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2472
|
+
return logf_lut(x);
|
|
2473
|
+
} else {
|
|
2474
|
+
const
|
|
2475
|
+
ln2_hi = reinterpret<f32>(0x3F317180), // 6.9313812256e-01f
|
|
2476
|
+
ln2_lo = reinterpret<f32>(0x3717F7D1), // 9.0580006145e-06f
|
|
2477
|
+
Lg1 = reinterpret<f32>(0x3F2AAAAA), // 0xaaaaaa.0p-24f
|
|
2478
|
+
Lg2 = reinterpret<f32>(0x3ECCCE13), // 0xccce13.0p-25f
|
|
2479
|
+
Lg3 = reinterpret<f32>(0x3E91E9EE), // 0x91e9ee.0p-25f
|
|
2480
|
+
Lg4 = reinterpret<f32>(0x3E789E26), // 0xf89e26.0p-26f
|
|
2481
|
+
Ox1p25f = reinterpret<f32>(0x4C000000);
|
|
2482
|
+
|
|
2483
|
+
let u = reinterpret<u32>(x);
|
|
2484
|
+
let k = 0;
|
|
2485
|
+
let sign = u >> 31;
|
|
2486
|
+
if (sign || u < 0x00800000) {
|
|
2487
|
+
if (u << 1 == 0) return -1 / (x * x);
|
|
2488
|
+
if (sign) return (x - x) / 0;
|
|
2489
|
+
k -= 25;
|
|
2490
|
+
x *= Ox1p25f;
|
|
2491
|
+
u = reinterpret<u32>(x);
|
|
2492
|
+
} else if (u >= 0x7F800000) {
|
|
2493
|
+
return x;
|
|
2494
|
+
} else if (u == 0x3F800000) {
|
|
2495
|
+
return 0;
|
|
2496
|
+
}
|
|
2497
|
+
u += 0x3F800000 - 0x3F3504F3;
|
|
2498
|
+
k += i32(u >> 23) - 0x7F;
|
|
2499
|
+
u = (u & 0x007FFFFF) + 0x3F3504F3;
|
|
2500
|
+
x = reinterpret<f32>(u);
|
|
2501
|
+
let f = x - 1.0;
|
|
2502
|
+
let s = f / (2.0 + f);
|
|
2503
|
+
let z = s * s;
|
|
2504
|
+
let w = z * z;
|
|
2505
|
+
let t1 = w * (Lg2 + w * Lg4);
|
|
2506
|
+
let t2 = z * (Lg1 + w * Lg3);
|
|
2507
|
+
let r = t2 + t1;
|
|
2508
|
+
let hfsq = <f32>0.5 * f * f;
|
|
2509
|
+
let dk = <f32>k;
|
|
2510
|
+
return s * (hfsq + r) + dk * ln2_lo - hfsq + f + dk * ln2_hi;
|
|
2511
|
+
}
|
|
2512
|
+
}
|
|
2513
|
+
|
|
2514
|
+
export function log10(x: f32): f32 { // see: musl/src/math/log10f.c and SUN COPYRIGHT NOTICE above
|
|
2515
|
+
const
|
|
2516
|
+
ivln10hi = reinterpret<f32>(0x3EDE6000), // 4.3432617188e-01f
|
|
2517
|
+
ivln10lo = reinterpret<f32>(0xB804EAD9), // -3.1689971365e-05f
|
|
2518
|
+
log10_2hi = reinterpret<f32>(0x3E9A2080), // 3.0102920532e-01f
|
|
2519
|
+
log10_2lo = reinterpret<f32>(0x355427DB), // 7.9034151668e-07f
|
|
2520
|
+
Lg1 = reinterpret<f32>(0x3F2AAAAA), // 0xaaaaaa.0p-24f, 0.66666662693f
|
|
2521
|
+
Lg2 = reinterpret<f32>(0x3ECCCE13), // 0xccce13.0p-25f, 0.40000972152f
|
|
2522
|
+
Lg3 = reinterpret<f32>(0x3E91E9EE), // 0x91e9ee.0p-25f, 0.28498786688f
|
|
2523
|
+
Lg4 = reinterpret<f32>(0x3E789E26), // 0xf89e26.0p-26f, 0.24279078841f
|
|
2524
|
+
Ox1p25f = reinterpret<f32>(0x4C000000); // 0x1p25f
|
|
2525
|
+
|
|
2526
|
+
let ux = reinterpret<u32>(x);
|
|
2527
|
+
let k = 0;
|
|
2528
|
+
let sign = ux >> 31;
|
|
2529
|
+
if (sign || ux < 0x00800000) {
|
|
2530
|
+
if (ux << 1 == 0) return -1 / (x * x);
|
|
2531
|
+
if (sign) return (x - x) / 0.0;
|
|
2532
|
+
k -= 25;
|
|
2533
|
+
x *= Ox1p25f;
|
|
2534
|
+
ux = reinterpret<u32>(x);
|
|
2535
|
+
} else if (ux >= 0x7F800000) {
|
|
2536
|
+
return x;
|
|
2537
|
+
} else if (ux == 0x3F800000) {
|
|
2538
|
+
return 0;
|
|
2539
|
+
}
|
|
2540
|
+
ux += 0x3F800000 - 0x3F3504F3;
|
|
2541
|
+
k += i32(ux >> 23) - 0x7F;
|
|
2542
|
+
ux = (ux & 0x007FFFFF) + 0x3F3504F3;
|
|
2543
|
+
x = reinterpret<f32>(ux);
|
|
2544
|
+
let f = x - 1.0;
|
|
2545
|
+
let s = f / (2.0 + f);
|
|
2546
|
+
let z = s * s;
|
|
2547
|
+
let w = z * z;
|
|
2548
|
+
let t1 = w * (Lg2 + w * Lg4);
|
|
2549
|
+
let t2 = z * (Lg1 + w * Lg3);
|
|
2550
|
+
let r = t2 + t1;
|
|
2551
|
+
let hfsq: f32 = 0.5 * f * f;
|
|
2552
|
+
let hi = f - hfsq;
|
|
2553
|
+
ux = reinterpret<u32>(hi);
|
|
2554
|
+
ux &= 0xFFFFF000;
|
|
2555
|
+
hi = reinterpret<f32>(ux);
|
|
2556
|
+
let lo = f - hi - hfsq + s * (hfsq + r);
|
|
2557
|
+
let dk = <f32>k;
|
|
2558
|
+
return dk * log10_2lo + (lo + hi) * ivln10lo + lo * ivln10hi + hi * ivln10hi + dk * log10_2hi;
|
|
2559
|
+
}
|
|
2560
|
+
|
|
2561
|
+
export function log1p(x: f32): f32 { // see: musl/src/math/log1pf.c and SUN COPYRIGHT NOTICE above
|
|
2562
|
+
const
|
|
2563
|
+
ln2_hi = reinterpret<f32>(0x3F317180), // 6.9313812256e-01
|
|
2564
|
+
ln2_lo = reinterpret<f32>(0x3717F7D1), // 9.0580006145e-06
|
|
2565
|
+
Lg1 = reinterpret<f32>(0x3F2AAAAA), // 0xaaaaaa.0p-24f, 0.66666662693f
|
|
2566
|
+
Lg2 = reinterpret<f32>(0x3ECCCE13), // 0xccce13.0p-25f, 0.40000972152f
|
|
2567
|
+
Lg3 = reinterpret<f32>(0x3E91E9EE), // 0x91e9ee.0p-25f, 0.28498786688f
|
|
2568
|
+
Lg4 = reinterpret<f32>(0x3E789E26); // 0xf89e26.0p-26f, 0.24279078841f
|
|
2569
|
+
|
|
2570
|
+
let ix = reinterpret<u32>(x);
|
|
2571
|
+
let c: f32 = 0;
|
|
2572
|
+
let f: f32 = 0;
|
|
2573
|
+
let k = 1;
|
|
2574
|
+
if (ix < 0x3ED413D0 || bool(ix >> 31)) {
|
|
2575
|
+
if (ix >= 0xBF800000) {
|
|
2576
|
+
if (x == -1) return x / 0.0;
|
|
2577
|
+
return (x - x) / 0.0;
|
|
2578
|
+
}
|
|
2579
|
+
if (ix << 1 < 0x33800000 << 1) return x;
|
|
2580
|
+
if (ix <= 0xBE95F619) {
|
|
2581
|
+
k = 0;
|
|
2582
|
+
c = 0;
|
|
2583
|
+
f = x;
|
|
2584
|
+
}
|
|
2585
|
+
} else if (ix >= 0x7F800000) return x;
|
|
2586
|
+
if (k) {
|
|
2587
|
+
let uf: f32 = 1 + x;
|
|
2588
|
+
let iu = reinterpret<u32>(uf);
|
|
2589
|
+
iu += 0x3F800000 - 0x3F3504F3;
|
|
2590
|
+
k = i32(iu >> 23) - 0x7F;
|
|
2591
|
+
if (k < 25) {
|
|
2592
|
+
c = k >= 2 ? 1 - (uf - x) : x - (uf - 1);
|
|
2593
|
+
c /= uf;
|
|
2594
|
+
} else c = 0;
|
|
2595
|
+
iu = (iu & 0x007FFFFF) + 0x3F3504F3;
|
|
2596
|
+
f = reinterpret<f32>(iu) - 1;
|
|
2597
|
+
}
|
|
2598
|
+
let s = f / (2.0 + f);
|
|
2599
|
+
let z = s * s;
|
|
2600
|
+
let w = z * z;
|
|
2601
|
+
let t1 = w * (Lg2 + w * Lg4);
|
|
2602
|
+
let t2 = z * (Lg1 + w * Lg3);
|
|
2603
|
+
let r = t2 + t1;
|
|
2604
|
+
let hfsq: f32 = 0.5 * f * f;
|
|
2605
|
+
let dk = <f32>k;
|
|
2606
|
+
return s * (hfsq + r) + (dk * ln2_lo + c) - hfsq + f + dk * ln2_hi;
|
|
2607
|
+
}
|
|
2608
|
+
|
|
2609
|
+
export function log2(x: f32): f32 { // see: musl/src/math/log2f.c and SUN COPYRIGHT NOTICE above
|
|
2610
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2611
|
+
return log2f_lut(x);
|
|
2612
|
+
} else {
|
|
2613
|
+
const
|
|
2614
|
+
ivln2hi = reinterpret<f32>(0x3FB8B000), // 1.4428710938e+00f
|
|
2615
|
+
ivln2lo = reinterpret<f32>(0xB9389AD4), // -1.7605285393e-04
|
|
2616
|
+
Lg1 = reinterpret<f32>(0x3F2AAAAA), // 0xaaaaaa.0p-24f, 0.66666662693f
|
|
2617
|
+
Lg2 = reinterpret<f32>(0x3ECCCE13), // 0xccce13.0p-25f, 0.40000972152f
|
|
2618
|
+
Lg3 = reinterpret<f32>(0x3E91E9EE), // 0x91e9ee.0p-25f, 0.28498786688f
|
|
2619
|
+
Lg4 = reinterpret<f32>(0x3E789E26), // 0xf89e26.0p-26f, 0.24279078841f
|
|
2620
|
+
Ox1p25f = reinterpret<f32>(0x4C000000); // 0x1p25f
|
|
2621
|
+
|
|
2622
|
+
let ux = reinterpret<u32>(x);
|
|
2623
|
+
let k = 0;
|
|
2624
|
+
let sign = ux >> 31;
|
|
2625
|
+
if (sign || ux < 0x00800000) {
|
|
2626
|
+
if (ux << 1 == 0) return -1 / (x * x);
|
|
2627
|
+
if (sign) return (x - x) / 0.0;
|
|
2628
|
+
k -= 25;
|
|
2629
|
+
x *= Ox1p25f;
|
|
2630
|
+
ux = reinterpret<u32>(x);
|
|
2631
|
+
} else if (ux >= 0x7F800000) {
|
|
2632
|
+
return x;
|
|
2633
|
+
} else if (ux == 0x3F800000) {
|
|
2634
|
+
return 0;
|
|
2635
|
+
}
|
|
2636
|
+
ux += 0x3F800000 - 0x3F3504F3;
|
|
2637
|
+
k += i32(ux >> 23) - 0x7F;
|
|
2638
|
+
ux = (ux & 0x007FFFFF) + 0x3F3504F3;
|
|
2639
|
+
x = reinterpret<f32>(ux);
|
|
2640
|
+
let f = x - 1.0;
|
|
2641
|
+
let s = f / (2.0 + f);
|
|
2642
|
+
let z = s * s;
|
|
2643
|
+
let w = z * z;
|
|
2644
|
+
let t1 = w * (Lg2 + w * Lg4);
|
|
2645
|
+
let t2 = z * (Lg1 + w * Lg3);
|
|
2646
|
+
let r = t2 + t1;
|
|
2647
|
+
let hfsq: f32 = 0.5 * f * f;
|
|
2648
|
+
let hi = f - hfsq;
|
|
2649
|
+
let u = reinterpret<u32>(hi);
|
|
2650
|
+
u &= 0xFFFFF000;
|
|
2651
|
+
hi = reinterpret<f32>(u);
|
|
2652
|
+
let lo: f32 = f - hi - hfsq + s * (hfsq + r);
|
|
2653
|
+
let dk = <f32>k;
|
|
2654
|
+
return (lo + hi) * ivln2lo + lo * ivln2hi + hi * ivln2hi + dk;
|
|
2655
|
+
}
|
|
2656
|
+
}
|
|
2657
|
+
|
|
2658
|
+
// @ts-ignore: decorator
|
|
2659
|
+
@inline
|
|
2660
|
+
export function max(value1: f32, value2: f32): f32 {
|
|
2661
|
+
return builtin_max<f32>(value1, value2);
|
|
2662
|
+
}
|
|
2663
|
+
|
|
2664
|
+
// @ts-ignore: decorator
|
|
2665
|
+
@inline
|
|
2666
|
+
export function min(value1: f32, value2: f32): f32 {
|
|
2667
|
+
return builtin_min<f32>(value1, value2);
|
|
2668
|
+
}
|
|
2669
|
+
|
|
2670
|
+
export function pow(x: f32, y: f32): f32 {
|
|
2671
|
+
// TODO: remove this fast pathes after introduced own mid-end IR with "stdlib call simplify" transforms
|
|
2672
|
+
if (builtin_abs<f32>(y) <= 2) {
|
|
2673
|
+
if (y == 2.0) return x * x;
|
|
2674
|
+
if (y == 0.5) {
|
|
2675
|
+
return select<f32>(
|
|
2676
|
+
builtin_abs<f32>(builtin_sqrt<f32>(x)),
|
|
2677
|
+
Infinity,
|
|
2678
|
+
x != -Infinity
|
|
2679
|
+
);
|
|
2680
|
+
}
|
|
2681
|
+
if (y == -1.0) return 1 / x;
|
|
2682
|
+
if (y == 1.0) return x;
|
|
2683
|
+
if (y == 0.0) return 1.0;
|
|
2684
|
+
}
|
|
2685
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2686
|
+
// see: musl/src/math/powf.c
|
|
2687
|
+
return powf_lut(x, y);
|
|
2688
|
+
} else {
|
|
2689
|
+
// based on: jdh8/metallic/src/math/float/powf.c
|
|
2690
|
+
if (y == 0) return 1;
|
|
2691
|
+
// @ts-ignore: cast
|
|
2692
|
+
if (isNaN(x) | isNaN(y)) {
|
|
2693
|
+
return NaN;
|
|
2694
|
+
}
|
|
2695
|
+
let sign: u32 = 0;
|
|
2696
|
+
let uy = reinterpret<u32>(y);
|
|
2697
|
+
let ux = reinterpret<u32>(x);
|
|
2698
|
+
let sx = ux >> 31;
|
|
2699
|
+
ux &= 0x7FFFFFFF;
|
|
2700
|
+
if (sx && nearest(y) == y) {
|
|
2701
|
+
x = -x;
|
|
2702
|
+
sx = 0;
|
|
2703
|
+
sign = u32(nearest(y * 0.5) != y * 0.5) << 31;
|
|
2704
|
+
}
|
|
2705
|
+
let m: u32;
|
|
2706
|
+
if (ux == 0x3F800000) { // x == 1
|
|
2707
|
+
m = sx | u32((uy & 0x7FFFFFFF) == 0x7F800000) ? 0x7FC00000 : 0x3F800000;
|
|
2708
|
+
} else if (ux == 0) {
|
|
2709
|
+
m = <i32>uy < 0 ? 0x7F800000 : 0;
|
|
2710
|
+
} else if (ux == 0x7F800000) {
|
|
2711
|
+
m = <i32>uy < 0 ? 0 : 0x7F800000;
|
|
2712
|
+
} else if (sx) {
|
|
2713
|
+
m = 0x7FC00000;
|
|
2714
|
+
} else {
|
|
2715
|
+
m = reinterpret<u32>(<f32>exp2f(<f64>y * log2f(x)));
|
|
2716
|
+
}
|
|
2717
|
+
return reinterpret<f32>(m | sign);
|
|
2718
|
+
}
|
|
2719
|
+
}
|
|
2720
|
+
|
|
2721
|
+
// @ts-ignore: decorator
|
|
2722
|
+
@inline
|
|
2723
|
+
export function seedRandom(value: i64): void {
|
|
2724
|
+
NativeMath.seedRandom(value);
|
|
2725
|
+
}
|
|
2726
|
+
|
|
2727
|
+
// Using xoroshiro64starstar from http://xoshiro.di.unimi.it/xoroshiro64starstar.c
|
|
2728
|
+
export function random(): f32 {
|
|
2729
|
+
if (!random_seeded) seedRandom(reinterpret<i64>(seed()));
|
|
2730
|
+
|
|
2731
|
+
let s0 = random_state0_32;
|
|
2732
|
+
let s1 = random_state1_32;
|
|
2733
|
+
let r = rotl<u32>(s0 * 0x9E3779BB, 5) * 5;
|
|
2734
|
+
|
|
2735
|
+
s1 ^= s0;
|
|
2736
|
+
random_state0_32 = rotl<u32>(s0, 26) ^ s1 ^ (s1 << 9);
|
|
2737
|
+
random_state1_32 = rotl<u32>(s1, 13);
|
|
2738
|
+
|
|
2739
|
+
return reinterpret<f32>((r >> 9) | (127 << 23)) - 1.0;
|
|
2740
|
+
}
|
|
2741
|
+
|
|
2742
|
+
export function round(x: f32): f32 {
|
|
2743
|
+
if (ASC_SHRINK_LEVEL > 0) {
|
|
2744
|
+
return builtin_ceil<f32>(x) - f32(builtin_ceil<f32>(x) - 0.5 > x);
|
|
2745
|
+
} else {
|
|
2746
|
+
let roundUp = builtin_ceil<f32>(x);
|
|
2747
|
+
return select<f32>(roundUp, roundUp - 1.0, roundUp - 0.5 <= x);
|
|
2748
|
+
}
|
|
2749
|
+
}
|
|
2750
|
+
|
|
2751
|
+
export function sign(x: f32): f32 {
|
|
2752
|
+
if (ASC_SHRINK_LEVEL > 0) {
|
|
2753
|
+
return select<f32>(builtin_copysign<f32>(1, x), x, builtin_abs(x) > 0);
|
|
2754
|
+
} else {
|
|
2755
|
+
return select<f32>(1, select<f32>(-1, x, x < 0), x > 0);
|
|
2756
|
+
}
|
|
2757
|
+
}
|
|
2758
|
+
|
|
2759
|
+
// @ts-ignore: decorator
|
|
2760
|
+
@inline
|
|
2761
|
+
export function signbit(x: f32): bool {
|
|
2762
|
+
return <bool>(reinterpret<u32>(x) >>> 31);
|
|
2763
|
+
}
|
|
2764
|
+
|
|
2765
|
+
export function sin(x: f32): f32 { // see: musl/src/math/sinf.c
|
|
2766
|
+
const
|
|
2767
|
+
s1pio2 = reinterpret<f64>(0x3FF921FB54442D18), // M_PI_2 * 1
|
|
2768
|
+
s2pio2 = reinterpret<f64>(0x400921FB54442D18), // M_PI_2 * 2
|
|
2769
|
+
s3pio2 = reinterpret<f64>(0x4012D97C7F3321D2), // M_PI_2 * 3
|
|
2770
|
+
s4pio2 = reinterpret<f64>(0x401921FB54442D18); // M_PI_2 * 4
|
|
2771
|
+
|
|
2772
|
+
let ux = reinterpret<u32>(x);
|
|
2773
|
+
let sign = ux >> 31;
|
|
2774
|
+
ux &= 0x7FFFFFFF;
|
|
2775
|
+
|
|
2776
|
+
if (ux <= 0x3F490FDA) { // |x| ~<= π/4
|
|
2777
|
+
if (ux < 0x39800000) { // |x| < 2**-12
|
|
2778
|
+
return x;
|
|
2779
|
+
}
|
|
2780
|
+
return sin_kernf(x);
|
|
2781
|
+
}
|
|
2782
|
+
|
|
2783
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2784
|
+
if (ux <= 0x407B53D1) { // |x| ~<= 5π/4
|
|
2785
|
+
if (ux <= 0x4016CBE3) { // |x| ~<= 3π/4
|
|
2786
|
+
return sign ? -cos_kernf(x + s1pio2) : cos_kernf(x - s1pio2);
|
|
2787
|
+
}
|
|
2788
|
+
return sin_kernf(-(sign ? x + s2pio2 : x - s2pio2));
|
|
2789
|
+
}
|
|
2790
|
+
|
|
2791
|
+
if (ux <= 0x40E231D5) { // |x| ~<= 9π/4
|
|
2792
|
+
if (ux <= 0x40AFEDDF) { // |x| ~<= 7π/4
|
|
2793
|
+
return sign ? cos_kernf(x + s3pio2) : -cos_kernf(x - s3pio2);
|
|
2794
|
+
}
|
|
2795
|
+
return sin_kernf(sign ? x + s4pio2 : x - s4pio2);
|
|
2796
|
+
}
|
|
2797
|
+
}
|
|
2798
|
+
|
|
2799
|
+
// sin(Inf or NaN) is NaN
|
|
2800
|
+
if (ux >= 0x7F800000) return x - x;
|
|
2801
|
+
|
|
2802
|
+
let n = rempio2f(x, ux, sign);
|
|
2803
|
+
let y = rempio2f_y;
|
|
2804
|
+
|
|
2805
|
+
let t = n & 1 ? cos_kernf(y) : sin_kernf(y);
|
|
2806
|
+
return n & 2 ? -t : t;
|
|
2807
|
+
}
|
|
2808
|
+
|
|
2809
|
+
export function sinh(x: f32): f32 { // see: musl/src/math/sinhf.c
|
|
2810
|
+
let u = reinterpret<u32>(x) & 0x7FFFFFFF;
|
|
2811
|
+
let a = reinterpret<f32>(u);
|
|
2812
|
+
let h = builtin_copysign<f32>(0.5, x);
|
|
2813
|
+
if (u < 0x42B17217) {
|
|
2814
|
+
let t = expm1(a);
|
|
2815
|
+
if (u < 0x3F800000) {
|
|
2816
|
+
if (u < 0x3F800000 - (12 << 23)) return x;
|
|
2817
|
+
return h * (2 * t - t * t / (t + 1));
|
|
2818
|
+
}
|
|
2819
|
+
return h * (t + t / (t + 1));
|
|
2820
|
+
}
|
|
2821
|
+
return expo2f(a, 2 * h);
|
|
2822
|
+
}
|
|
2823
|
+
|
|
2824
|
+
// @ts-ignore: decorator
|
|
2825
|
+
@inline
|
|
2826
|
+
export function sqrt(x: f32): f32 {
|
|
2827
|
+
return builtin_sqrt<f32>(x);
|
|
2828
|
+
}
|
|
2829
|
+
|
|
2830
|
+
export function tan(x: f32): f32 { // see: musl/src/math/tanf.c
|
|
2831
|
+
const
|
|
2832
|
+
t1pio2 = reinterpret<f64>(0x3FF921FB54442D18), // 1 * M_PI_2
|
|
2833
|
+
t2pio2 = reinterpret<f64>(0x400921FB54442D18), // 2 * M_PI_2
|
|
2834
|
+
t3pio2 = reinterpret<f64>(0x4012D97C7F3321D2), // 3 * M_PI_2
|
|
2835
|
+
t4pio2 = reinterpret<f64>(0x401921FB54442D18); // 4 * M_PI_2
|
|
2836
|
+
|
|
2837
|
+
let ux = reinterpret<u32>(x);
|
|
2838
|
+
let sign = ux >> 31;
|
|
2839
|
+
ux &= 0x7FFFFFFF;
|
|
2840
|
+
|
|
2841
|
+
if (ux <= 0x3F490FDA) { // |x| ~<= π/4
|
|
2842
|
+
if (ux < 0x39800000) { // |x| < 2**-12
|
|
2843
|
+
return x;
|
|
2844
|
+
}
|
|
2845
|
+
return tan_kernf(x, 0);
|
|
2846
|
+
}
|
|
2847
|
+
|
|
2848
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
2849
|
+
if (ux <= 0x407B53D1) { // |x| ~<= 5π/4
|
|
2850
|
+
if (ux <= 0x4016CBE3) { // |x| ~<= 3π/4
|
|
2851
|
+
return tan_kernf((sign ? x + t1pio2 : x - t1pio2), 1);
|
|
2852
|
+
} else {
|
|
2853
|
+
return tan_kernf((sign ? x + t2pio2 : x - t2pio2), 0);
|
|
2854
|
+
}
|
|
2855
|
+
}
|
|
2856
|
+
if (ux <= 0x40E231D5) { // |x| ~<= 9π/4
|
|
2857
|
+
if (ux <= 0x40AFEDDF) { // |x| ~<= 7π/4
|
|
2858
|
+
return tan_kernf((sign ? x + t3pio2 : x - t3pio2), 1);
|
|
2859
|
+
} else {
|
|
2860
|
+
return tan_kernf((sign ? x + t4pio2 : x - t4pio2), 0);
|
|
2861
|
+
}
|
|
2862
|
+
}
|
|
2863
|
+
}
|
|
2864
|
+
|
|
2865
|
+
// tan(Inf or NaN) is NaN
|
|
2866
|
+
if (ux >= 0x7F800000) return x - x;
|
|
2867
|
+
|
|
2868
|
+
// argument reduction
|
|
2869
|
+
let n = rempio2f(x, ux, sign);
|
|
2870
|
+
let y = rempio2f_y;
|
|
2871
|
+
return tan_kernf(y, n & 1);
|
|
2872
|
+
}
|
|
2873
|
+
|
|
2874
|
+
export function tanh(x: f32): f32 { // see: musl/src/math/tanhf.c
|
|
2875
|
+
let u = reinterpret<u32>(x);
|
|
2876
|
+
u &= 0x7FFFFFFF;
|
|
2877
|
+
let y = reinterpret<f32>(u);
|
|
2878
|
+
let t: f32;
|
|
2879
|
+
if (u > 0x3F0C9F54) {
|
|
2880
|
+
if (u > 0x41200000) t = 1 + 0 / y;
|
|
2881
|
+
else {
|
|
2882
|
+
t = expm1(2 * y);
|
|
2883
|
+
t = 1 - 2 / (t + 2);
|
|
2884
|
+
}
|
|
2885
|
+
} else if (u > 0x3E82C578) {
|
|
2886
|
+
t = expm1(2 * y);
|
|
2887
|
+
t = t / (t + 2);
|
|
2888
|
+
} else if (u >= 0x00800000) {
|
|
2889
|
+
t = expm1(-2 * y);
|
|
2890
|
+
t = -t / (t + 2);
|
|
2891
|
+
} else t = y;
|
|
2892
|
+
return builtin_copysign<f32>(t, x);
|
|
2893
|
+
}
|
|
2894
|
+
|
|
2895
|
+
// @ts-ignore: decorator
|
|
2896
|
+
@inline
|
|
2897
|
+
export function trunc(x: f32): f32 {
|
|
2898
|
+
return builtin_trunc<f32>(x);
|
|
2899
|
+
}
|
|
2900
|
+
|
|
2901
|
+
export function scalbn(x: f32, n: i32): f32 { // see: https://git.musl-libc.org/cgit/musl/tree/src/math/scalbnf.c
|
|
2902
|
+
const
|
|
2903
|
+
Ox1p24f = reinterpret<f32>(0x4B800000),
|
|
2904
|
+
Ox1p127f = reinterpret<f32>(0x7F000000),
|
|
2905
|
+
Ox1p_126f = reinterpret<f32>(0x00800000);
|
|
2906
|
+
|
|
2907
|
+
let y = x;
|
|
2908
|
+
if (n > 127) {
|
|
2909
|
+
y *= Ox1p127f;
|
|
2910
|
+
n -= 127;
|
|
2911
|
+
if (n > 127) {
|
|
2912
|
+
y *= Ox1p127f;
|
|
2913
|
+
n = builtin_min<i32>(n - 127, 127);
|
|
2914
|
+
}
|
|
2915
|
+
} else if (n < -126) {
|
|
2916
|
+
y *= Ox1p_126f * Ox1p24f;
|
|
2917
|
+
n += 126 - 24;
|
|
2918
|
+
if (n < -126) {
|
|
2919
|
+
y *= Ox1p_126f * Ox1p24f;
|
|
2920
|
+
n = builtin_max<i32>(n + 126 - 24, -126);
|
|
2921
|
+
}
|
|
2922
|
+
}
|
|
2923
|
+
return y * reinterpret<f32>(<u32>(0x7F + n) << 23);
|
|
2924
|
+
}
|
|
2925
|
+
|
|
2926
|
+
export function mod(x: f32, y: f32): f32 { // see: musl/src/math/fmodf.c
|
|
2927
|
+
if (builtin_abs<f32>(y) == 1.0) {
|
|
2928
|
+
// x % 1, x % -1 ==> sign(x) * abs(x - 1.0 * trunc(x / 1.0))
|
|
2929
|
+
// TODO: move this rule to compiler's optimization pass.
|
|
2930
|
+
// It could be apply for any x % C_pot, where "C_pot" is pow of two const.
|
|
2931
|
+
return builtin_copysign<f32>(x - builtin_trunc<f32>(x), x);
|
|
2932
|
+
}
|
|
2933
|
+
let ux = reinterpret<u32>(x);
|
|
2934
|
+
let uy = reinterpret<u32>(y);
|
|
2935
|
+
let ex = i32(ux >> 23 & 0xFF);
|
|
2936
|
+
let ey = i32(uy >> 23 & 0xFF);
|
|
2937
|
+
let sm = ux & 0x80000000;
|
|
2938
|
+
let uy1 = uy << 1;
|
|
2939
|
+
if (uy1 == 0 || ex == 0xFF || isNaN<f32>(y)) {
|
|
2940
|
+
let m = x * y;
|
|
2941
|
+
return m / m;
|
|
2942
|
+
}
|
|
2943
|
+
let ux1 = ux << 1;
|
|
2944
|
+
if (ux1 <= uy1) {
|
|
2945
|
+
return x * f32(ux1 != uy1);
|
|
2946
|
+
}
|
|
2947
|
+
if (!ex) {
|
|
2948
|
+
ex -= builtin_clz<u32>(ux << 9);
|
|
2949
|
+
ux <<= 1 - ex;
|
|
2950
|
+
} else {
|
|
2951
|
+
ux &= <u32>-1 >> 9;
|
|
2952
|
+
ux |= 1 << 23;
|
|
2953
|
+
}
|
|
2954
|
+
if (!ey) {
|
|
2955
|
+
ey -= builtin_clz<u32>(uy << 9);
|
|
2956
|
+
uy <<= 1 - ey;
|
|
2957
|
+
} else {
|
|
2958
|
+
uy &= u32(-1) >> 9;
|
|
2959
|
+
uy |= 1 << 23;
|
|
2960
|
+
}
|
|
2961
|
+
while (ex > ey) {
|
|
2962
|
+
if (ux >= uy) {
|
|
2963
|
+
if (ux == uy) return 0 * x;
|
|
2964
|
+
ux -= uy;
|
|
2965
|
+
}
|
|
2966
|
+
ux <<= 1;
|
|
2967
|
+
--ex;
|
|
2968
|
+
}
|
|
2969
|
+
if (ux >= uy) {
|
|
2970
|
+
if (ux == uy) return 0 * x;
|
|
2971
|
+
ux -= uy;
|
|
2972
|
+
}
|
|
2973
|
+
// for (; !(ux >> 23); ux <<= 1) --ex;
|
|
2974
|
+
let shift = <i32>builtin_clz<u32>(ux << 8);
|
|
2975
|
+
ex -= shift;
|
|
2976
|
+
ux <<= shift;
|
|
2977
|
+
if (ex > 0) {
|
|
2978
|
+
ux -= 1 << 23;
|
|
2979
|
+
ux |= <u32>ex << 23;
|
|
2980
|
+
} else {
|
|
2981
|
+
ux >>= -ex + 1;
|
|
2982
|
+
}
|
|
2983
|
+
return reinterpret<f32>(ux | sm);
|
|
2984
|
+
}
|
|
2985
|
+
|
|
2986
|
+
export function rem(x: f32, y: f32): f32 { // see: musl/src/math/remquof.c
|
|
2987
|
+
let ux = reinterpret<u32>(x);
|
|
2988
|
+
let uy = reinterpret<u32>(y);
|
|
2989
|
+
let ex = i32(ux >> 23 & 0xFF);
|
|
2990
|
+
let ey = i32(uy >> 23 & 0xFF);
|
|
2991
|
+
let uxi = ux;
|
|
2992
|
+
if (uy << 1 == 0 || ex == 0xFF || isNaN(y)) return (x * y) / (x * y);
|
|
2993
|
+
if (ux << 1 == 0) return x;
|
|
2994
|
+
if (!ex) {
|
|
2995
|
+
ex -= builtin_clz<u32>(uxi << 9);
|
|
2996
|
+
uxi <<= 1 - ex;
|
|
2997
|
+
} else {
|
|
2998
|
+
uxi &= u32(-1) >> 9;
|
|
2999
|
+
uxi |= 1 << 23;
|
|
3000
|
+
}
|
|
3001
|
+
if (!ey) {
|
|
3002
|
+
ey -= builtin_clz<u32>(uy << 9);
|
|
3003
|
+
uy <<= 1 - ey;
|
|
3004
|
+
} else {
|
|
3005
|
+
uy &= u32(-1) >> 9;
|
|
3006
|
+
uy |= 1 << 23;
|
|
3007
|
+
}
|
|
3008
|
+
let q = 0;
|
|
3009
|
+
do {
|
|
3010
|
+
if (ex < ey) {
|
|
3011
|
+
if (ex + 1 == ey) break; // goto end
|
|
3012
|
+
return x;
|
|
3013
|
+
}
|
|
3014
|
+
while (ex > ey) {
|
|
3015
|
+
if (uxi >= uy) {
|
|
3016
|
+
uxi -= uy;
|
|
3017
|
+
++q;
|
|
3018
|
+
}
|
|
3019
|
+
uxi <<= 1;
|
|
3020
|
+
q <<= 1;
|
|
3021
|
+
--ex;
|
|
3022
|
+
}
|
|
3023
|
+
if (uxi >= uy) {
|
|
3024
|
+
uxi -= uy;
|
|
3025
|
+
++q;
|
|
3026
|
+
}
|
|
3027
|
+
if (uxi == 0) ex = -30;
|
|
3028
|
+
else {
|
|
3029
|
+
let shift = builtin_clz<i32>(uxi << 8);
|
|
3030
|
+
ex -= shift;
|
|
3031
|
+
uxi <<= shift;
|
|
3032
|
+
}
|
|
3033
|
+
break;
|
|
3034
|
+
} while (false);
|
|
3035
|
+
// end:
|
|
3036
|
+
if (ex > 0) {
|
|
3037
|
+
uxi -= 1 << 23;
|
|
3038
|
+
uxi |= <u32>ex << 23;
|
|
3039
|
+
} else {
|
|
3040
|
+
uxi >>= -ex + 1;
|
|
3041
|
+
}
|
|
3042
|
+
x = reinterpret<f32>(uxi);
|
|
3043
|
+
y = builtin_abs<f32>(y);
|
|
3044
|
+
let x2 = x + x;
|
|
3045
|
+
if (ex == ey || (ex + 1 == ey && (<f32>x2 > y || (<f32>x2 == y && bool(q & 1))))) {
|
|
3046
|
+
x -= y;
|
|
3047
|
+
// q++;
|
|
3048
|
+
}
|
|
3049
|
+
return <i32>ux < 0 ? -x : x;
|
|
3050
|
+
}
|
|
3051
|
+
|
|
3052
|
+
export function sincos(x: f32): void { // see: musl/tree/src/math/sincosf.c
|
|
3053
|
+
const
|
|
3054
|
+
s1pio2 = reinterpret<f64>(0x3FF921FB54442D18), // 1 * M_PI_2
|
|
3055
|
+
s2pio2 = reinterpret<f64>(0x400921FB54442D18), // 2 * M_PI_2
|
|
3056
|
+
s3pio2 = reinterpret<f64>(0x4012D97C7F3321D2), // 3 * M_PI_2
|
|
3057
|
+
s4pio2 = reinterpret<f64>(0x401921FB54442D18); // 4 * M_PI_2
|
|
3058
|
+
|
|
3059
|
+
let ux = reinterpret<u32>(x);
|
|
3060
|
+
let sign = ux >> 31;
|
|
3061
|
+
ux &= 0x7FFFFFFF;
|
|
3062
|
+
|
|
3063
|
+
if (ux <= 0x3F490FDA) { // |x| ~<= π/4
|
|
3064
|
+
if (ux < 0x39800000) { // |x| < 2**-12
|
|
3065
|
+
sincos_sin = x;
|
|
3066
|
+
sincos_cos = 1;
|
|
3067
|
+
return;
|
|
3068
|
+
}
|
|
3069
|
+
sincos_sin = sin_kernf(x);
|
|
3070
|
+
sincos_cos = cos_kernf(x);
|
|
3071
|
+
return;
|
|
3072
|
+
}
|
|
3073
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
3074
|
+
if (ux <= 0x407B53D1) { // |x| ~<= 5π/4
|
|
3075
|
+
if (ux <= 0x4016CBE3) { // |x| ~<= 3π/4
|
|
3076
|
+
if (sign) {
|
|
3077
|
+
sincos_sin = -cos_kernf(x + s1pio2);
|
|
3078
|
+
sincos_cos = sin_kernf(x + s1pio2);
|
|
3079
|
+
} else {
|
|
3080
|
+
sincos_sin = cos_kernf(s1pio2 - x);
|
|
3081
|
+
sincos_cos = sin_kernf(s1pio2 - x);
|
|
3082
|
+
}
|
|
3083
|
+
return;
|
|
3084
|
+
}
|
|
3085
|
+
// -sin(x + c) is not correct if x+c could be 0: -0 vs +0
|
|
3086
|
+
sincos_sin = -sin_kernf(sign ? x + s2pio2 : x - s2pio2);
|
|
3087
|
+
sincos_cos = -cos_kernf(sign ? x + s2pio2 : x - s2pio2);
|
|
3088
|
+
return;
|
|
3089
|
+
}
|
|
3090
|
+
if (ux <= 0x40E231D5) { // |x| ~<= 9π/4
|
|
3091
|
+
if (ux <= 0x40AFEDDF) { // |x| ~<= 7π/4
|
|
3092
|
+
if (sign) {
|
|
3093
|
+
sincos_sin = cos_kernf(x + s3pio2);
|
|
3094
|
+
sincos_cos = -sin_kernf(x + s3pio2);
|
|
3095
|
+
} else {
|
|
3096
|
+
sincos_sin = -cos_kernf(x - s3pio2);
|
|
3097
|
+
sincos_cos = sin_kernf(x - s3pio2);
|
|
3098
|
+
}
|
|
3099
|
+
return;
|
|
3100
|
+
}
|
|
3101
|
+
sincos_sin = sin_kernf(sign ? x + s4pio2 : x - s4pio2);
|
|
3102
|
+
sincos_cos = cos_kernf(sign ? x + s4pio2 : x - s4pio2);
|
|
3103
|
+
return;
|
|
3104
|
+
}
|
|
3105
|
+
}
|
|
3106
|
+
// sin(Inf or NaN) is NaN
|
|
3107
|
+
if (ux >= 0x7F800000) {
|
|
3108
|
+
let xx = x - x;
|
|
3109
|
+
sincos_sin = xx;
|
|
3110
|
+
sincos_cos = xx;
|
|
3111
|
+
return;
|
|
3112
|
+
}
|
|
3113
|
+
// general argument reduction needed
|
|
3114
|
+
let n = rempio2f(x, ux, sign);
|
|
3115
|
+
let y = rempio2f_y;
|
|
3116
|
+
let s = sin_kernf(y);
|
|
3117
|
+
let c = cos_kernf(y);
|
|
3118
|
+
let sin = s, cos = c;
|
|
3119
|
+
if (n & 1) {
|
|
3120
|
+
sin = c;
|
|
3121
|
+
cos = -s;
|
|
3122
|
+
}
|
|
3123
|
+
if (n & 2) {
|
|
3124
|
+
sin = -sin;
|
|
3125
|
+
cos = -cos;
|
|
3126
|
+
}
|
|
3127
|
+
sincos_sin = sin;
|
|
3128
|
+
sincos_cos = cos;
|
|
3129
|
+
}
|
|
3130
|
+
}
|
|
3131
|
+
|
|
3132
|
+
export function ipow32(x: i32, e: i32): i32 {
|
|
3133
|
+
let out = 1;
|
|
3134
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
3135
|
+
if (x == 2) {
|
|
3136
|
+
return select<i32>(1 << e, 0, <u32>e < 32);
|
|
3137
|
+
}
|
|
3138
|
+
if (e <= 0) {
|
|
3139
|
+
if (x == -1) return select<i32>(-1, 1, e & 1);
|
|
3140
|
+
return i32(e == 0) | i32(x == 1);
|
|
3141
|
+
}
|
|
3142
|
+
else if (e == 1) return x;
|
|
3143
|
+
else if (e == 2) return x * x;
|
|
3144
|
+
else if (e < 32) {
|
|
3145
|
+
let log = 32 - clz(e);
|
|
3146
|
+
// 32 = 2 ^ 5, so need only five cases.
|
|
3147
|
+
// But some extra cases needs for properly overflowing
|
|
3148
|
+
switch (log) {
|
|
3149
|
+
case 5: {
|
|
3150
|
+
if (e & 1) out *= x;
|
|
3151
|
+
e >>>= 1;
|
|
3152
|
+
x *= x;
|
|
3153
|
+
}
|
|
3154
|
+
case 4: {
|
|
3155
|
+
if (e & 1) out *= x;
|
|
3156
|
+
e >>>= 1;
|
|
3157
|
+
x *= x;
|
|
3158
|
+
}
|
|
3159
|
+
case 3: {
|
|
3160
|
+
if (e & 1) out *= x;
|
|
3161
|
+
e >>>= 1;
|
|
3162
|
+
x *= x;
|
|
3163
|
+
}
|
|
3164
|
+
case 2: {
|
|
3165
|
+
if (e & 1) out *= x;
|
|
3166
|
+
e >>>= 1;
|
|
3167
|
+
x *= x;
|
|
3168
|
+
}
|
|
3169
|
+
case 1: {
|
|
3170
|
+
if (e & 1) out *= x;
|
|
3171
|
+
}
|
|
3172
|
+
}
|
|
3173
|
+
return out;
|
|
3174
|
+
}
|
|
3175
|
+
}
|
|
3176
|
+
while (e) {
|
|
3177
|
+
if (e & 1) out *= x;
|
|
3178
|
+
e >>>= 1;
|
|
3179
|
+
x *= x;
|
|
3180
|
+
}
|
|
3181
|
+
return out;
|
|
3182
|
+
}
|
|
3183
|
+
|
|
3184
|
+
export function ipow64(x: i64, e: i64): i64 {
|
|
3185
|
+
let out: i64 = 1;
|
|
3186
|
+
if (ASC_SHRINK_LEVEL < 1) {
|
|
3187
|
+
if (x == 2) {
|
|
3188
|
+
return select<i64>(1 << e, 0, <u64>e < 64);
|
|
3189
|
+
}
|
|
3190
|
+
if (e <= 0) {
|
|
3191
|
+
if (x == -1) return select<i64>(-1, 1, e & 1);
|
|
3192
|
+
return i64(e == 0) | i64(x == 1);
|
|
3193
|
+
}
|
|
3194
|
+
else if (e == 1) return x;
|
|
3195
|
+
else if (e == 2) return x * x;
|
|
3196
|
+
else if (e < 64) {
|
|
3197
|
+
let log = 64 - <i32>clz(e);
|
|
3198
|
+
// 64 = 2 ^ 6, so need only six cases.
|
|
3199
|
+
// But some extra cases needs for properly overflowing
|
|
3200
|
+
switch (log) {
|
|
3201
|
+
case 6: {
|
|
3202
|
+
if (e & 1) out *= x;
|
|
3203
|
+
e >>>= 1;
|
|
3204
|
+
x *= x;
|
|
3205
|
+
}
|
|
3206
|
+
case 5: {
|
|
3207
|
+
if (e & 1) out *= x;
|
|
3208
|
+
e >>>= 1;
|
|
3209
|
+
x *= x;
|
|
3210
|
+
}
|
|
3211
|
+
case 4: {
|
|
3212
|
+
if (e & 1) out *= x;
|
|
3213
|
+
e >>>= 1;
|
|
3214
|
+
x *= x;
|
|
3215
|
+
}
|
|
3216
|
+
case 3: {
|
|
3217
|
+
if (e & 1) out *= x;
|
|
3218
|
+
e >>>= 1;
|
|
3219
|
+
x *= x;
|
|
3220
|
+
}
|
|
3221
|
+
case 2: {
|
|
3222
|
+
if (e & 1) out *= x;
|
|
3223
|
+
e >>>= 1;
|
|
3224
|
+
x *= x;
|
|
3225
|
+
}
|
|
3226
|
+
case 1: {
|
|
3227
|
+
if (e & 1) out *= x;
|
|
3228
|
+
}
|
|
3229
|
+
}
|
|
3230
|
+
return out;
|
|
3231
|
+
}
|
|
3232
|
+
}
|
|
3233
|
+
while (e) {
|
|
3234
|
+
if (e & 1) out *= x;
|
|
3235
|
+
e >>>= 1;
|
|
3236
|
+
x *= x;
|
|
3237
|
+
}
|
|
3238
|
+
return out;
|
|
3239
|
+
}
|
|
3240
|
+
|
|
3241
|
+
/*
|
|
3242
|
+
TODO:
|
|
3243
|
+
In compile time if only exponent is constant we could replace ipow32/ipow64 by shortest addition chains
|
|
3244
|
+
which usually faster than exponentiation by squaring
|
|
3245
|
+
|
|
3246
|
+
for ipow32 and e < 32:
|
|
3247
|
+
|
|
3248
|
+
let b: i32, c: i32, d: i32, h: i32, k: i32, g: i32;
|
|
3249
|
+
switch (e) {
|
|
3250
|
+
case 1: return x;
|
|
3251
|
+
case 2: return x * x;
|
|
3252
|
+
case 3: return x * x * x;
|
|
3253
|
+
case 4: return (b = x * x) * b;
|
|
3254
|
+
case 5: return (b = x * x) * b * x;
|
|
3255
|
+
case 6: return (b = x * x) * b * b;
|
|
3256
|
+
case 7: return (b = x * x) * b * b * x;
|
|
3257
|
+
case 8: return (d = (b = x * x) * b) * d;
|
|
3258
|
+
case 9: return (c = x * x * x) * c * c;
|
|
3259
|
+
case 10: return (d = (b = x * x) * b) * d * b;
|
|
3260
|
+
case 11: return (d = (b = x * x) * b) * d * b * x;
|
|
3261
|
+
case 12: return (d = (b = x * x) * b) * d * d;
|
|
3262
|
+
case 13: return (d = (b = x * x) * b) * d * d * x;
|
|
3263
|
+
case 14: return (d = (b = x * x) * b) * d * d * b;
|
|
3264
|
+
case 15: return (k = (b = x * x) * b * x) * k * k;
|
|
3265
|
+
case 16: return (h = (d = (b = x * x) * b) * d) * h;
|
|
3266
|
+
case 17: return (h = (d = (b = x * x) * b) * d) * h * x;
|
|
3267
|
+
case 18: return (h = (d = (b = x * x) * b) * d * x) * h;
|
|
3268
|
+
case 19: return (h = (d = (b = x * x) * b) * d * x) * h * x;
|
|
3269
|
+
case 20: return (h = (k = (b = x * x) * b * x) * k) * h;
|
|
3270
|
+
case 21: return (h = (k = (b = x * x) * b * x) * k) * h * x;
|
|
3271
|
+
case 22: return (g = (h = (k = (b = x * x) * b * x) * k) * x) * g;
|
|
3272
|
+
case 23: return (h = (d = (c = (b = x * x) * x) * b) * d) * h * c;
|
|
3273
|
+
case 24: return (h = (d = (c = x * x * x) * c) * d) * h;
|
|
3274
|
+
case 25: return (h = (d = (c = x * x * x) * c) * d) * h * x;
|
|
3275
|
+
case 26: return (g = (h = (d = (c = x * x * x) * c) * d) * x) * g;
|
|
3276
|
+
case 27: return (h = (d = (c = x * x * x) * c) * d) * h * c;
|
|
3277
|
+
case 28: return (h = (d = (c = x * x * x) * c * x) * d) * h;
|
|
3278
|
+
case 29: return (h = (d = (c = x * x * x) * c * x) * d) * h * x;
|
|
3279
|
+
case 30: return (h = (d = (c = x * x * x) * c) * d * c) * h;
|
|
3280
|
+
case 31: return (h = (d = (c = x * x * x) * c) * d * c) * h * x;
|
|
3281
|
+
}
|
|
3282
|
+
|
|
3283
|
+
for ipow64: TODO
|
|
3284
|
+
switch (e) {
|
|
3285
|
+
case 32:
|
|
3286
|
+
...
|
|
3287
|
+
case 63:
|
|
3288
|
+
}
|
|
3289
|
+
*/
|