|
| 1 | +/* ef_sqrtf.c -- float version of e_sqrt.c. |
| 2 | + * Conversion to float by Ian Lance Taylor, Cygnus Support, [email protected]. |
| 3 | + */ |
| 4 | + |
| 5 | +/* |
| 6 | + * ==================================================== |
| 7 | + * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
| 8 | + * |
| 9 | + * Developed at SunPro, a Sun Microsystems, Inc. business. |
| 10 | + * Permission to use, copy, modify, and distribute this |
| 11 | + * software is freely granted, provided that this notice |
| 12 | + * is preserved. |
| 13 | + * ==================================================== |
| 14 | + */ |
| 15 | + |
| 16 | +const ONE: f32 = 1.0; |
| 17 | +const TINY: f32 = 1.0e-30; |
| 18 | + |
| 19 | +#[inline] |
| 20 | +#[cfg_attr(all(test, assert_no_panic), no_panic::no_panic)] |
| 21 | +pub fn sqrtf(x: f32) -> f32 { |
| 22 | + use super::fdlibm::{FLT_UWORD_IS_FINITE, FLT_UWORD_IS_SUBNORMAL, FLT_UWORD_IS_ZERO}; |
| 23 | + // On wasm32 we know that LLVM's intrinsic will compile to an optimized |
| 24 | + // `f32.sqrt` native instruction, so we can leverage this for both code size |
| 25 | + // and speed. |
| 26 | + llvm_intrinsically_optimized! { |
| 27 | + #[cfg(target_arch = "wasm32")] { |
| 28 | + return if x < 0.0 { |
| 29 | + ::core::f32::NAN |
| 30 | + } else { |
| 31 | + unsafe { ::core::intrinsics::sqrtf32(x) } |
| 32 | + } |
| 33 | + } |
| 34 | + } |
| 35 | + |
| 36 | + let mut z: f32; |
| 37 | + |
| 38 | + let mut r: u32; |
| 39 | + let hx: u32; |
| 40 | + |
| 41 | + let mut ix: i32; |
| 42 | + let mut s: i32; |
| 43 | + let mut q: i32; |
| 44 | + let mut m: i32; |
| 45 | + let mut t: i32; |
| 46 | + let mut i: i32; |
| 47 | + |
| 48 | + ix = x.to_bits() as i32; |
| 49 | + hx = ix as u32 & 0x7fff_ffff; |
| 50 | + |
| 51 | + /* take care of Inf and NaN */ |
| 52 | + if !FLT_UWORD_IS_FINITE(hx) { |
| 53 | + return x * x + x; /* sqrt(NaN)=NaN, sqrt(+inf)=+inf, sqrt(-inf)=sNaN */ |
| 54 | + } |
| 55 | + |
| 56 | + /* take care of zero and -ves */ |
| 57 | + if FLT_UWORD_IS_ZERO(hx) { |
| 58 | + return x; /* sqrt(+-0) = +-0 */ |
| 59 | + } |
| 60 | + if ix < 0 { |
| 61 | + return (x - x) / (x - x); /* sqrt(-ve) = sNaN */ |
| 62 | + } |
| 63 | + |
| 64 | + /* normalize x */ |
| 65 | + m = ix >> 23; |
| 66 | + if FLT_UWORD_IS_SUBNORMAL(hx) { |
| 67 | + /* subnormal x */ |
| 68 | + i = 0; |
| 69 | + while ix & 0x0080_0000 == 0 { |
| 70 | + ix <<= 1; |
| 71 | + i += 1; |
| 72 | + } |
| 73 | + m -= i - 1; |
| 74 | + } |
| 75 | + m -= 127; /* unbias exponent */ |
| 76 | + ix = (ix & 0x007f_ffff) | 0x0080_0000; |
| 77 | + /* odd m, double x to make it even */ |
| 78 | + if m & 1 == 1 { |
| 79 | + ix += ix; |
| 80 | + } |
| 81 | + m >>= 1; /* m = [m/2] */ |
| 82 | + |
| 83 | + /* generate sqrt(x) bit by bit */ |
| 84 | + ix += ix; |
| 85 | + q = 0; |
| 86 | + s = 0; /* q = sqrt(x) */ |
| 87 | + r = 0x0100_0000; /* r = moving bit from right to left */ |
| 88 | + |
| 89 | + while r != 0 { |
| 90 | + t = s + r as i32; |
| 91 | + if t <= ix { |
| 92 | + s = t + r as i32; |
| 93 | + ix -= t; |
| 94 | + q += r as i32; |
| 95 | + } |
| 96 | + ix += ix; |
| 97 | + r >>= 1; |
| 98 | + } |
| 99 | + |
| 100 | + /* use floating add to find out rounding direction */ |
| 101 | + if ix != 0 { |
| 102 | + z = ONE - TINY; /* trigger inexact flag */ |
| 103 | + if z >= ONE { |
| 104 | + z = ONE + TINY; |
| 105 | + if z > ONE { |
| 106 | + q += 2; |
| 107 | + } else { |
| 108 | + q += q & 1; |
| 109 | + } |
| 110 | + } |
| 111 | + } |
| 112 | + ix = (q >> 1) + 0x3f00_0000; |
| 113 | + ix += m << 23; |
| 114 | + f32::from_bits(ix as u32) |
| 115 | +} |
0 commit comments