mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-05-23 05:42:29 +00:00
Use ARM's faster math functions on non-tiny builds
This commit is contained in:
Justine Tunney
parent
3c10fb5580
commit
4814b6bdf8
58 changed files with 3760 additions and 361 deletions
151
libc/tinymath/log.c
Normal file
151
libc/tinymath/log.c
Normal file
|
|
@ -0,0 +1,151 @@
|
|||
/*-*- mode:c;indent-tabs-mode:t;c-basic-offset:8;tab-width:8;coding:utf-8 -*-│
|
||||
│vi: set et ft=c ts=8 tw=8 fenc=utf-8 :vi│
|
||||
╚──────────────────────────────────────────────────────────────────────────────╝
|
||||
│ │
|
||||
│ Musl Libc │
|
||||
│ Copyright © 2005-2014 Rich Felker, et al. │
|
||||
│ │
|
||||
│ Permission is hereby granted, free of charge, to any person obtaining │
|
||||
│ a copy of this software and associated documentation files (the │
|
||||
│ "Software"), to deal in the Software without restriction, including │
|
||||
│ without limitation the rights to use, copy, modify, merge, publish, │
|
||||
│ distribute, sublicense, and/or sell copies of the Software, and to │
|
||||
│ permit persons to whom the Software is furnished to do so, subject to │
|
||||
│ the following conditions: │
|
||||
│ │
|
||||
│ The above copyright notice and this permission notice shall be │
|
||||
│ included in all copies or substantial portions of the Software. │
|
||||
│ │
|
||||
│ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, │
|
||||
│ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF │
|
||||
│ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. │
|
||||
│ IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY │
|
||||
│ CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, │
|
||||
│ TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE │
|
||||
│ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. │
|
||||
│ │
|
||||
╚─────────────────────────────────────────────────────────────────────────────*/
|
||||
#include "libc/bits/likely.h"
|
||||
#include "libc/math.h"
|
||||
#include "libc/tinymath/internal.h"
|
||||
#include "libc/tinymath/log_data.internal.h"
|
||||
#ifndef TINY
|
||||
|
||||
asm(".ident\t\"\\n\\n\
|
||||
Double-precision math functions (MIT License)\\n\
|
||||
Copyright 2018 ARM Limited\"");
|
||||
asm(".include \"libc/disclaimer.inc\"");
|
||||
/* clang-format off */
|
||||
|
||||
/*
|
||||
* Double-precision log(x) function.
|
||||
*
|
||||
* Copyright (c) 2018, Arm Limited.
|
||||
* SPDX-License-Identifier: MIT
|
||||
*/
|
||||
|
||||
#define T __log_data.tab
|
||||
#define T2 __log_data.tab2
|
||||
#define B __log_data.poly1
|
||||
#define A __log_data.poly
|
||||
#define Ln2hi __log_data.ln2hi
|
||||
#define Ln2lo __log_data.ln2lo
|
||||
#define N (1 << LOG_TABLE_BITS)
|
||||
#define OFF 0x3fe6000000000000
|
||||
|
||||
/* Top 16 bits of a double. */
|
||||
static inline uint32_t top16(double x)
|
||||
{
|
||||
return asuint64(x) >> 48;
|
||||
}
|
||||
|
||||
/**
|
||||
* Returns natural logarithm of 𝑥.
|
||||
*/
|
||||
double log(double x)
|
||||
{
|
||||
double_t w, z, r, r2, r3, y, invc, logc, kd, hi, lo;
|
||||
uint64_t ix, iz, tmp;
|
||||
uint32_t top;
|
||||
int k, i;
|
||||
|
||||
ix = asuint64(x);
|
||||
top = top16(x);
|
||||
#define LO asuint64(1.0 - 0x1p-4)
|
||||
#define HI asuint64(1.0 + 0x1.09p-4)
|
||||
if (UNLIKELY(ix - LO < HI - LO)) {
|
||||
/* Handle close to 1.0 inputs separately. */
|
||||
/* Fix sign of zero with downward rounding when x==1. */
|
||||
if (WANT_ROUNDING && UNLIKELY(ix == asuint64(1.0)))
|
||||
return 0;
|
||||
r = x - 1.0;
|
||||
r2 = r * r;
|
||||
r3 = r * r2;
|
||||
y = r3 *
|
||||
(B[1] + r * B[2] + r2 * B[3] +
|
||||
r3 * (B[4] + r * B[5] + r2 * B[6] +
|
||||
r3 * (B[7] + r * B[8] + r2 * B[9] + r3 * B[10])));
|
||||
/* Worst-case error is around 0.507 ULP. */
|
||||
w = r * 0x1p27;
|
||||
double_t rhi = r + w - w;
|
||||
double_t rlo = r - rhi;
|
||||
w = rhi * rhi * B[0]; /* B[0] == -0.5. */
|
||||
hi = r + w;
|
||||
lo = r - hi + w;
|
||||
lo += B[0] * rlo * (rhi + r);
|
||||
y += lo;
|
||||
y += hi;
|
||||
return eval_as_double(y);
|
||||
}
|
||||
if (UNLIKELY(top - 0x0010 >= 0x7ff0 - 0x0010)) {
|
||||
/* x < 0x1p-1022 or inf or nan. */
|
||||
if (ix * 2 == 0)
|
||||
return __math_divzero(1);
|
||||
if (ix == asuint64(INFINITY)) /* log(inf) == inf. */
|
||||
return x;
|
||||
if ((top & 0x8000) || (top & 0x7ff0) == 0x7ff0)
|
||||
return __math_invalid(x);
|
||||
/* x is subnormal, normalize it. */
|
||||
ix = asuint64(x * 0x1p52);
|
||||
ix -= 52ULL << 52;
|
||||
}
|
||||
|
||||
/* x = 2^k z; where z is in range [OFF,2*OFF) and exact.
|
||||
The range is split into N subintervals.
|
||||
The ith subinterval contains z and c is near its center. */
|
||||
tmp = ix - OFF;
|
||||
i = (tmp >> (52 - LOG_TABLE_BITS)) % N;
|
||||
k = (int64_t)tmp >> 52; /* arithmetic shift */
|
||||
iz = ix - (tmp & 0xfffULL << 52);
|
||||
invc = T[i].invc;
|
||||
logc = T[i].logc;
|
||||
z = asdouble(iz);
|
||||
|
||||
/* log(x) = log1p(z/c-1) + log(c) + k*Ln2. */
|
||||
/* r ~= z/c - 1, |r| < 1/(2*N). */
|
||||
#if __FP_FAST_FMA
|
||||
/* rounding error: 0x1p-55/N. */
|
||||
r = __builtin_fma(z, invc, -1.0);
|
||||
#else
|
||||
/* rounding error: 0x1p-55/N + 0x1p-66. */
|
||||
r = (z - T2[i].chi - T2[i].clo) * invc;
|
||||
#endif
|
||||
kd = (double_t)k;
|
||||
|
||||
/* hi + lo = r + log(c) + k*Ln2. */
|
||||
w = kd * Ln2hi + logc;
|
||||
hi = w + r;
|
||||
lo = w - hi + r + kd * Ln2lo;
|
||||
|
||||
/* log(x) = lo + (log1p(r) - r) + hi. */
|
||||
r2 = r * r; /* rounding error: 0x1p-54/N^2. */
|
||||
/* Worst case error if |y| > 0x1p-5:
|
||||
0.5 + 4.13/N + abs-poly-error*2^57 ULP (+ 0.002 ULP without fma)
|
||||
Worst case error if |y| > 0x1p-4:
|
||||
0.5 + 2.06/N + abs-poly-error*2^56 ULP (+ 0.001 ULP without fma). */
|
||||
y = lo + r2 * A[0] +
|
||||
r * r2 * (A[1] + r * A[2] + r2 * (A[3] + r * A[4])) + hi;
|
||||
return eval_as_double(y);
|
||||
}
|
||||
|
||||
#endif /* !TINY */
|
||||
Loading…
Add table
Add a link
Reference in a new issue