Add libm to mikros std

This commit is contained in:
pjht 2024-08-06 17:31:19 -05:00
parent aefc4ee221
commit a70304b471
Signed by: pjht
GPG Key ID: 7B5F6AFBEC7EE78E
137 changed files with 14329 additions and 1 deletions

View File

@ -1787,7 +1787,7 @@ version = "2.1.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6cb51c9a029ddc91b07a787f1d86b53ccfa49b0e86688c946ebe8d3555685dd7"
dependencies = [
"libm",
"libm 0.2.8 (registry+https://github.com/rust-lang/crates.io-index)",
]
[[package]]
@ -2258,6 +2258,14 @@ dependencies = [
"windows-targets 0.48.5",
]
[[package]]
name = "libm"
version = "0.2.8"
dependencies = [
"compiler_builtins",
"rustc-std-workspace-core",
]
[[package]]
name = "libm"
version = "0.2.8"
@ -5418,6 +5426,7 @@ dependencies = [
"hashbrown",
"hermit-abi 0.4.0",
"libc",
"libm 0.2.8",
"miniz_oxide",
"object 0.36.0",
"panic_abort",

View File

@ -65,6 +65,7 @@ x86_64 = { path = "../../mikros_std_deps/x86_64-0.15.1", default-features = fals
elf = { path = "../../mikros_std_deps/elf-0.7.4/", default-features = false, features = ["rustc-dep-of-std"] }
serde = { path = "../../mikros_std_deps/serde-1.0.203/", default-features = false, features = ["rustc-dep-of-std", "alloc"]}
postcard = { path = "../../mikros_std_deps/postcard-1.0.8/", default-features = false, features = ["rustc-dep-of-std", "alloc"]}
libm = { path = "../../mikros_std_deps/libm-0.2.8/", default-features = false, features = ["rustc-dep-of-std"] }
[features]

View File

@ -0,0 +1,570 @@
#[no_mangle]
#[allow(unused)]
pub extern "C" fn acosf(x: f32) -> f32 {
libm::acosf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn acoshf(x: f32) -> f32 {
libm::acoshf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn asinf(x: f32) -> f32 {
libm::asinf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn asinhf(x: f32) -> f32 {
libm::asinhf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atanf(x: f32) -> f32 {
libm::atanf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atan2f(y: f32, x: f32) -> f32 {
libm::atan2f(y, x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atanhf(x: f32) -> f32 {
libm::atanhf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn cbrtf(x: f32) -> f32 {
libm::cbrtf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ceilf(x: f32) -> f32 {
libm::ceilf(x)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn copysignf(x: f32, y: f32) -> f32 {
// libm::copysignf(x, y)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn cosf(x: f32) -> f32 {
libm::cosf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn coshf(x: f32) -> f32 {
libm::coshf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn erff(x: f32) -> f32 {
libm::erff(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn erfcf(x: f32) -> f32 {
libm::erfcf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn expf(x: f32) -> f32 {
libm::expf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn exp2f(x: f32) -> f32 {
libm::exp2f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn exp10f(x: f32) -> f32 {
libm::exp10f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn expm1f(x: f32) -> f32 {
libm::expm1f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fabsf(x: f32) -> f32 {
libm::fabsf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fdimf(x: f32, y: f32) -> f32 {
libm::fdimf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn floorf(x: f32) -> f32 {
libm::floorf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmaf(x: f32, y: f32, z: f32) -> f32 {
libm::fmaf(x, y, z)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmaxf(x: f32, y: f32) -> f32 {
libm::fmaxf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fminf(x: f32, y: f32) -> f32 {
libm::fminf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmodf(x: f32, y: f32) -> f32 {
libm::fmodf(x, y)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn frexpf(x: f32) -> (f32, i32) {
// libm::frexpf(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn hypotf(x: f32, y: f32) -> f32 {
libm::hypotf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ilogbf(x: f32) -> i32 {
libm::ilogbf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn j0f(x: f32) -> f32 {
libm::j0f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn j1f(x: f32) -> f32 {
libm::j1f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn jnf(n: i32, x: f32) -> f32 {
libm::jnf(n, x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ldexpf(x: f32, n: i32) -> f32 {
libm::ldexpf(x, n)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn lgammaf_r(x: f32) -> (f32, i32) {
// libm::lgammaf_r(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn lgammaf(x: f32) -> f32 {
libm::lgammaf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn logf(x: f32) -> f32 {
libm::logf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log1pf(x: f32) -> f32 {
libm::log1pf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log2f(x: f32) -> f32 {
libm::log2f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log10f(x: f32) -> f32 {
libm::log10f(x)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn modff(x: f32) -> (f32, f32) {
// libm::modff(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn nextafterf(x: f32, y: f32) -> f32 {
libm::nextafterf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn powf(x: f32, y: f32) -> f32 {
libm::powf(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn remainderf(x: f32, y: f32) -> f32 {
libm::remainderf(x, y)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn remquof(x: f32, y: f32) -> (f32, i32) {
// libm::remquof(x, y)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn rintf(x: f32) -> f32 {
libm::rintf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn roundf(x: f32) -> f32 {
libm::roundf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn scalbnf(x: f32, n: i32) -> f32 {
libm::scalbnf(x, n)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sinf(x: f32) -> f32 {
libm::sinf(x)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn sincosf(x: f32) -> (f32, f32) {
// libm::sincosf(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sinhf(x: f32) -> f32 {
libm::sinhf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sqrtf(x: f32) -> f32 {
libm::sqrtf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tanf(x: f32) -> f32 {
libm::tanf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tanhf(x: f32) -> f32 {
libm::tanhf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tgammaf(x: f32) -> f32 {
libm::tgammaf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn truncf(x: f32) -> f32 {
libm::truncf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn y0f(x: f32) -> f32 {
libm::y0f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn y1f(x: f32) -> f32 {
libm::y1f(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ynf(n: i32, x: f32) -> f32 {
libm::ynf(n, x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn acos(x: f64) -> f64 {
libm::acos(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn acosh(x: f64) -> f64 {
libm::acosh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn asin(x: f64) -> f64 {
libm::asin(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn asinh(x: f64) -> f64 {
libm::asinh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atan(x: f64) -> f64 {
libm::atan(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atan2(y: f64, x: f64) -> f64 {
libm::atan2(y, x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn atanh(x: f64) -> f64 {
libm::atanh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn cbrt(x: f64) -> f64 {
libm::cbrt(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ceil(x: f64) -> f64 {
libm::ceil(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn copysign(x: f64, y: f64) -> f64 {
libm::copysign(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn cos(x: f64) -> f64 {
libm::cos(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn cosh(x: f64) -> f64 {
libm::cosh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn erf(x: f64) -> f64 {
libm::erf(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn erfc(x: f64) -> f64 {
libm::erfc(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn exp(x: f64) -> f64 {
libm::exp(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn exp2(x: f64) -> f64 {
libm::exp2(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn exp10(x: f64) -> f64 {
libm::exp10(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn expm1(x: f64) -> f64 {
libm::expm1(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fabs(x: f64) -> f64 {
libm::fabs(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fdim(x: f64, y: f64) -> f64 {
libm::fdim(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn floor(x: f64) -> f64 {
libm::floor(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fma(x: f64, y: f64, z: f64) -> f64 {
libm::fma(x, y, z)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmax(x: f64, y: f64) -> f64 {
libm::fmax(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmin(x: f64, y: f64) -> f64 {
libm::fmin(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn fmod(x: f64, y: f64) -> f64 {
libm::fmod(x, y)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn frexp(x: f64) -> (f64, i32) {
// libm::frexp(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn hypot(x: f64, y: f64) -> f64 {
libm::hypot(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ilogb(x: f64) -> i32 {
libm::ilogb(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn j0(x: f64) -> f64 {
libm::j0(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn j1(x: f64) -> f64 {
libm::j1(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn jn(n: i32, x: f64) -> f64 {
libm::jn(n, x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn ldexp(x: f64, n: i32) -> f64 {
libm::ldexp(x, n)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn lgamma_r(x: f64) -> (f64, i32) {
// libm::lgamma_r(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn lgamma(x: f64) -> f64 {
libm::lgamma(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log(x: f64) -> f64 {
libm::log(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log1p(x: f64) -> f64 {
libm::log1p(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log2(x: f64) -> f64 {
libm::log2(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn log10(x: f64) -> f64 {
libm::log10(x)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn modf(x: f64) -> (f64, f64) {
// libm::modf(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn nextafter(x: f64, y: f64) -> f64 {
libm::nextafter(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn pow(x: f64, y: f64) -> f64 {
libm::pow(x, y)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn remainder(x: f64, y: f64) -> f64 {
libm::remainder(x, y)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn remquo(x: f64, y: f64) -> (f64, i32) {
// libm::remquo(x, y)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn rint(x: f64) -> f64 {
libm::rint(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn round(x: f64) -> f64 {
libm::round(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn scalbn(x: f64, n: i32) -> f64 {
libm::scalbn(x, n)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sin(x: f64) -> f64 {
libm::sin(x)
}
//#[no_mangle]
//#[allow(unused)]
//pub extern "C" fn sincos(x: f64) -> (f64, f64) {
// libm::sincos(x)
//}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sinh(x: f64) -> f64 {
libm::sinh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn sqrt(x: f64) -> f64 {
libm::sqrt(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tan(x: f64) -> f64 {
libm::tan(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tanh(x: f64) -> f64 {
libm::tanh(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn tgamma(x: f64) -> f64 {
libm::tgamma(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn trunc(x: f64) -> f64 {
libm::trunc(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn y0(x: f64) -> f64 {
libm::y0(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn y1(x: f64) -> f64 {
libm::y1(x)
}
#[no_mangle]
#[allow(unused)]
pub extern "C" fn yn(n: i32, x: f64) -> f64 {
libm::yn(n, x)
}

View File

@ -23,3 +23,5 @@
mod common;
pub use common::*;
mod math;

View File

@ -0,0 +1,6 @@
{
"git": {
"sha1": "721a5edc1be6b0412e4b1704590aed76f9a55899"
},
"path_in_vcs": ""
}

View File

@ -0,0 +1,21 @@
# EditorConfig helps developers define and maintain consistent
# coding styles between different editors and IDEs
# editorconfig.org
root = true
[*]
end_of_line = lf
charset = utf-8
trim_trailing_whitespace = true
insert_final_newline = true
indent_style = space
indent_size = 4
[*.md]
# double whitespace at end of line
# denotes a line break in Markdown
trim_trailing_whitespace = false
[*.yml]
indent_size = 2

8
mikros_std_deps/libm-0.2.8/.gitignore vendored Normal file
View File

@ -0,0 +1,8 @@
**/*.rs.bk
.#*
/bin
/math/src
/math/target
/target
/tests
Cargo.lock

View File

@ -0,0 +1,123 @@
# Change Log
All notable changes to this project will be documented in this file.
This project adheres to [Semantic Versioning](http://semver.org/).
## [Unreleased]
...
## [v0.2.1] - 2019-11-22
### Fixed
- sincosf
## [v0.2.0] - 2019-10-18
### Added
- Benchmarks
- signum
- remainder
- remainderf
- nextafter
- nextafterf
### Fixed
- Rounding to negative zero
- Overflows in rem_pio2 and remquo
- Overflows in fma
- sincosf
### Removed
- F32Ext and F64Ext traits
## [v0.1.4] - 2019-06-12
### Fixed
- Restored compatibility with Rust 1.31.0
## [v0.1.3] - 2019-05-14
### Added
- minf
- fmin
- fmaxf
- fmax
## [v0.1.2] - 2018-07-18
### Added
- acosf
- asin
- asinf
- atan
- atan2
- atan2f
- atanf
- cos
- cosf
- cosh
- coshf
- exp2
- expm1
- expm1f
- expo2
- fmaf
- pow
- sin
- sinf
- sinh
- sinhf
- tan
- tanf
- tanh
- tanhf
## [v0.1.1] - 2018-07-14
### Added
- acos
- acosf
- asin
- asinf
- atanf
- cbrt
- cbrtf
- ceil
- ceilf
- cosf
- exp
- exp2
- exp2f
- expm1
- expm1f
- fdim
- fdimf
- floorf
- fma
- fmod
- log
- log2
- log10
- log10f
- log1p
- log1pf
- log2f
- roundf
- sinf
- tanf
## v0.1.0 - 2018-07-13
- Initial release
[Unreleased]: https://github.com/japaric/libm/compare/v0.2.1...HEAD
[v0.2.1]: https://github.com/japaric/libm/compare/0.2.0...v0.2.1
[v0.2.0]: https://github.com/japaric/libm/compare/0.1.4...v0.2.0
[v0.1.4]: https://github.com/japaric/libm/compare/0.1.3...v0.1.4
[v0.1.3]: https://github.com/japaric/libm/compare/v0.1.2...0.1.3
[v0.1.2]: https://github.com/japaric/libm/compare/v0.1.1...v0.1.2
[v0.1.1]: https://github.com/japaric/libm/compare/v0.1.0...v0.1.1

View File

@ -0,0 +1,95 @@
# How to contribute
- Pick your favorite math function from the [issue tracker].
- Look for the C implementation of the function in the [MUSL source code][src].
- Copy paste the C code into a Rust file in the `src/math` directory and adjust
`src/math/mod.rs` accordingly. Also, uncomment the corresponding trait method
in `src/lib.rs`.
- Write some simple tests in your module (using `#[test]`)
- Run `cargo test` to make sure it works
- Run `cargo test --features musl-reference-tests` to compare your
implementation against musl's
- Send us a pull request! Make sure to run `cargo fmt` on your code before
sending the PR. Also include "closes #42" in the PR description to close the
corresponding issue.
- :tada:
[issue tracker]: https://github.com/rust-lang/libm/issues
[src]: https://git.musl-libc.org/cgit/musl/tree/src/math
[`src/math/truncf.rs`]: https://github.com/rust-lang/libm/blob/master/src/math/truncf.rs
Check [PR #65] for an example.
[PR #65]: https://github.com/rust-lang/libm/pull/65
## Tips and tricks
- *IMPORTANT* The code in this crate will end up being used in the `core` crate so it can **not**
have any external dependencies (other than `core` itself).
- Only use relative imports within the `math` directory / module, e.g. `use self::fabs::fabs` or
`use super::k_cos`. Absolute imports from core are OK, e.g. `use core::u64`.
- To reinterpret a float as an integer use the `to_bits` method. The MUSL code uses the
`GET_FLOAT_WORD` macro, or a union, to do this operation.
- To reinterpret an integer as a float use the `f32::from_bits` constructor. The MUSL code uses the
`SET_FLOAT_WORD` macro, or a union, to do this operation.
- You may use other methods from core like `f64::is_nan`, etc. as appropriate.
- If you're implementing one of the private double-underscore functions, take a look at the
"source" name in the comment at the top for an idea for alternate naming. For example, `__sin`
was renamed to `k_sin` after the FreeBSD source code naming. Do `use` these private functions in
`mod.rs`.
- You may encounter weird literals like `0x1p127f` in the MUSL code. These are hexadecimal floating
point literals. Rust (the language) doesn't support these kind of literals. The best way I have
found to deal with these literals is to turn them into their integer representation using the
[`hexf!`] macro and then turn them back into floats. See below:
[`hexf!`]: https://crates.io/crates/hexf
``` rust
// Step 1: write a program to convert the float into its integer representation
#[macro_use]
extern crate hexf;
fn main() {
println!("{:#x}", hexf32!("0x1.0p127").to_bits());
}
```
``` console
$ # Step 2: run the program
$ cargo run
0x7f000000
```
``` rust
// Step 3: copy paste the output into libm
let x1p127 = f32::from_bits(0x7f000000); // 0x1p127f === 2 ^ 12
```
- Rust code panics on arithmetic overflows when not optimized. You may need to use the [`Wrapping`]
newtype to avoid this problem.
[`Wrapping`]: https://doc.rust-lang.org/std/num/struct.Wrapping.html
## Testing
Normal tests can be executed with:
```
cargo test
```
If you'd like to run tests with randomized inputs that get compared against musl
itself, you'll need to be on a Linux system and then you can execute:
```
cargo test --features musl-reference-tests
```
Note that you may need to pass `--release` to Cargo if there are errors related
to integer overflow.

View File

@ -0,0 +1,59 @@
# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO
#
# When uploading crates to the registry Cargo will automatically
# "normalize" Cargo.toml files for maximal compatibility
# with all versions of Cargo and also rewrite `path` dependencies
# to registry (e.g., crates.io) dependencies.
#
# If you are reading this file be aware that the original Cargo.toml
# will likely look very different (and much more reasonable).
# See Cargo.toml.orig for the original contents.
[package]
edition = "2018"
name = "libm"
version = "0.2.8"
authors = ["Jorge Aparicio <jorge@japaric.io>"]
exclude = [
"/ci/",
"/.github/workflows/",
]
description = "libm in pure Rust"
documentation = "https://docs.rs/libm"
readme = "README.md"
keywords = [
"libm",
"math",
]
categories = ["no-std"]
license = "MIT OR Apache-2.0"
repository = "https://github.com/rust-lang/libm"
[profile.release]
lto = "fat"
[dev-dependencies.no-panic]
version = "0.1.8"
[build-dependencies.rand]
version = "0.6.5"
optional = true
[dependencies.core]
version = "1.0.0"
optional = true
package = "rustc-std-workspace-core"
[dependencies.compiler_builtins]
version = "0.1"
optional = true
[features]
default = []
musl-reference-tests = ["rand"]
unstable = []
rustc-dep-of-std = [
"core",
"compiler_builtins",
]

View File

@ -0,0 +1,40 @@
[package]
authors = ["Jorge Aparicio <jorge@japaric.io>"]
categories = ["no-std"]
description = "libm in pure Rust"
documentation = "https://docs.rs/libm"
keywords = ["libm", "math"]
license = "MIT OR Apache-2.0"
name = "libm"
readme = "README.md"
repository = "https://github.com/rust-lang/libm"
version = "0.2.8"
edition = "2018"
exclude = ["/ci/", "/.github/workflows/"]
[features]
default = []
# This tells the compiler to assume that a Nightly toolchain is being used and
# that it should activate any useful Nightly things accordingly.
unstable = []
# Generate tests which are random inputs and the outputs are calculated with
# musl libc.
musl-reference-tests = ['rand']
[workspace]
members = [
"crates/compiler-builtins-smoke-test",
"crates/libm-bench",
]
[dev-dependencies]
no-panic = "0.1.8"
[build-dependencies]
rand = { version = "0.6.5", optional = true }
# This is needed for no-panic to correctly detect the lack of panics
[profile.release]
lto = "fat"

View File

@ -0,0 +1,201 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

View File

@ -0,0 +1,25 @@
Copyright (c) 2018 Jorge Aparicio
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.

View File

@ -0,0 +1,51 @@
# `libm`
A port of [MUSL]'s libm to Rust.
[MUSL]: https://musl.libc.org/
## Goals
The short term goal of this library is to [enable math support (e.g. `sin`, `atan2`) for the
`wasm32-unknown-unknown` target][wasm] (cf. [rust-lang/compiler-builtins][pr]). The longer
term goal is to enable [math support in the `core` crate][core].
[wasm]: https://github.com/rust-lang/libm/milestone/1
[pr]: https://github.com/rust-lang/compiler-builtins/pull/248
[core]: https://github.com/rust-lang/libm/milestone/2
## Already usable
This crate is [on crates.io] and can be used today in stable `#![no_std]` programs.
The API documentation can be found [here](https://docs.rs/libm).
[on crates.io]: https://crates.io/crates/libm
## Benchmark
[benchmark]: #benchmark
The benchmarks are located in `crates/libm-bench` and require a nightly Rust toolchain.
To run all benchmarks:
> cargo +nightly bench --all
## Contributing
Please check [CONTRIBUTING.md](CONTRIBUTING.md)
## License
Licensed under either of
- Apache License, Version 2.0 ([LICENSE-APACHE](LICENSE-APACHE) or
http://www.apache.org/licenses/LICENSE-2.0)
- MIT license ([LICENSE-MIT](LICENSE-MIT) or http://opensource.org/licenses/MIT)
at your option.
### Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the
work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any
additional terms or conditions.

View File

@ -0,0 +1,444 @@
//use std::env;
fn main() {
println!("cargo:rerun-if-changed=build.rs");
#[cfg(feature = "musl-reference-tests")]
musl_reference_tests::generate();
//if !cfg!(feature = "checked") {
// let lvl = env::var("OPT_LEVEL").unwrap();
// if lvl != "0" {
// println!("cargo:rustc-cfg=assert_no_panic");
// }
//}
}
#[cfg(feature = "musl-reference-tests")]
mod musl_reference_tests {
use rand::seq::SliceRandom;
use rand::Rng;
use std::env;
use std::fs;
use std::process::Command;
// Number of tests to generate for each function
const NTESTS: usize = 500;
// These files are all internal functions or otherwise miscellaneous, not
// defining a function we want to test.
const IGNORED_FILES: &[&str] = &[
"fenv.rs",
// These are giving slightly different results compared to musl
"lgamma.rs",
"lgammaf.rs",
"tgamma.rs",
"j0.rs",
"j0f.rs",
"jn.rs",
"jnf.rs",
"j1.rs",
"j1f.rs",
];
struct Function {
name: String,
args: Vec<Ty>,
ret: Vec<Ty>,
tests: Vec<Test>,
}
enum Ty {
F32,
F64,
I32,
Bool,
}
struct Test {
inputs: Vec<i64>,
outputs: Vec<i64>,
}
pub fn generate() {
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
let target_arch = env::var("CARGO_CFG_TARGET_ARCH").unwrap();
if target_arch == "powerpc64" {
return;
}
let files =
fs::read_dir("src/math").unwrap().map(|f| f.unwrap().path()).collect::<Vec<_>>();
let mut math = Vec::new();
for file in files {
if IGNORED_FILES.iter().any(|f| file.ends_with(f)) {
continue;
}
println!("generating musl reference tests in {:?}", file);
let contents = fs::read_to_string(file).unwrap();
let mut functions = contents.lines().filter(|f| f.starts_with("pub fn"));
while let Some(function_to_test) = functions.next() {
math.push(parse(function_to_test));
}
}
// Generate a bunch of random inputs for each function. This will
// attempt to generate a good set of uniform test cases for exercising
// all the various functionality.
generate_random_tests(&mut math, &mut rand::thread_rng());
// After we have all our inputs, use the x86_64-unknown-linux-musl
// target to generate the expected output.
generate_test_outputs(&mut math);
//panic!("Boo");
// ... and now that we have both inputs and expected outputs, do a bunch
// of codegen to create the unit tests which we'll actually execute.
generate_unit_tests(&math);
}
/// A "poor man's" parser for the signature of a function
fn parse(s: &str) -> Function {
let s = eat(s, "pub fn ");
let pos = s.find('(').unwrap();
let name = &s[..pos];
let s = &s[pos + 1..];
let end = s.find(')').unwrap();
let args = s[..end]
.split(',')
.map(|arg| {
let colon = arg.find(':').unwrap();
parse_ty(arg[colon + 1..].trim())
})
.collect::<Vec<_>>();
let tail = &s[end + 1..];
let tail = eat(tail, " -> ");
let ret = parse_retty(tail.replace("{", "").trim());
return Function { name: name.to_string(), args, ret, tests: Vec::new() };
fn parse_ty(s: &str) -> Ty {
match s {
"f32" => Ty::F32,
"f64" => Ty::F64,
"i32" => Ty::I32,
"bool" => Ty::Bool,
other => panic!("unknown type `{}`", other),
}
}
fn parse_retty(s: &str) -> Vec<Ty> {
match s {
"(f32, f32)" => vec![Ty::F32, Ty::F32],
"(f32, i32)" => vec![Ty::F32, Ty::I32],
"(f64, f64)" => vec![Ty::F64, Ty::F64],
"(f64, i32)" => vec![Ty::F64, Ty::I32],
other => vec![parse_ty(other)],
}
}
fn eat<'a>(s: &'a str, prefix: &str) -> &'a str {
if s.starts_with(prefix) {
&s[prefix.len()..]
} else {
panic!("{:?} didn't start with {:?}", s, prefix)
}
}
}
fn generate_random_tests<R: Rng>(functions: &mut [Function], rng: &mut R) {
for function in functions {
for _ in 0..NTESTS {
function.tests.push(generate_test(function, rng));
}
}
fn generate_test<R: Rng>(function: &Function, rng: &mut R) -> Test {
let mut inputs = function.args.iter().map(|ty| ty.gen_i64(rng)).collect::<Vec<_>>();
// First argument to this function appears to be a number of
// iterations, so passing in massive random numbers causes it to
// take forever to execute, so make sure we're not running random
// math code until the heat death of the universe.
if function.name == "jn" || function.name == "jnf" {
inputs[0] &= 0xffff;
}
Test {
inputs,
// zero output for now since we'll generate it later
outputs: vec![],
}
}
}
impl Ty {
fn gen_i64<R: Rng>(&self, r: &mut R) -> i64 {
use std::f32;
use std::f64;
return match self {
Ty::F32 => {
if r.gen_range(0, 20) < 1 {
let i = *[f32::NAN, f32::INFINITY, f32::NEG_INFINITY].choose(r).unwrap();
i.to_bits().into()
} else {
r.gen::<f32>().to_bits().into()
}
}
Ty::F64 => {
if r.gen_range(0, 20) < 1 {
let i = *[f64::NAN, f64::INFINITY, f64::NEG_INFINITY].choose(r).unwrap();
i.to_bits() as i64
} else {
r.gen::<f64>().to_bits() as i64
}
}
Ty::I32 => {
if r.gen_range(0, 10) < 1 {
let i = *[i32::max_value(), 0, i32::min_value()].choose(r).unwrap();
i.into()
} else {
r.gen::<i32>().into()
}
}
Ty::Bool => r.gen::<bool>() as i64,
};
}
fn libc_ty(&self) -> &'static str {
match self {
Ty::F32 => "f32",
Ty::F64 => "f64",
Ty::I32 => "i32",
Ty::Bool => "i32",
}
}
fn libc_pty(&self) -> &'static str {
match self {
Ty::F32 => "*mut f32",
Ty::F64 => "*mut f64",
Ty::I32 => "*mut i32",
Ty::Bool => "*mut i32",
}
}
fn default(&self) -> &'static str {
match self {
Ty::F32 => "0_f32",
Ty::F64 => "0_f64",
Ty::I32 => "0_i32",
Ty::Bool => "false",
}
}
fn to_i64(&self) -> &'static str {
match self {
Ty::F32 => ".to_bits() as i64",
Ty::F64 => ".to_bits() as i64",
Ty::I32 => " as i64",
Ty::Bool => " as i64",
}
}
}
fn generate_test_outputs(functions: &mut [Function]) {
let mut src = String::new();
let dst = std::env::var("OUT_DIR").unwrap();
// Generate a program which will run all tests with all inputs in
// `functions`. This program will write all outputs to stdout (in a
// binary format).
src.push_str("use std::io::Write;");
src.push_str("fn main() {");
src.push_str("let mut result = Vec::new();");
for function in functions.iter_mut() {
src.push_str("unsafe {");
src.push_str("extern { fn ");
src.push_str(&function.name);
src.push_str("(");
let (ret, retptr) = match function.name.as_str() {
"sincos" | "sincosf" => (None, &function.ret[..]),
_ => (Some(&function.ret[0]), &function.ret[1..]),
};
for (i, arg) in function.args.iter().enumerate() {
src.push_str(&format!("arg{}: {},", i, arg.libc_ty()));
}
for (i, ret) in retptr.iter().enumerate() {
src.push_str(&format!("argret{}: {},", i, ret.libc_pty()));
}
src.push_str(")");
if let Some(ty) = ret {
src.push_str(" -> ");
src.push_str(ty.libc_ty());
}
src.push_str("; }");
src.push_str(&format!("static TESTS: &[[i64; {}]]", function.args.len()));
src.push_str(" = &[");
for test in function.tests.iter() {
src.push_str("[");
for val in test.inputs.iter() {
src.push_str(&val.to_string());
src.push_str(",");
}
src.push_str("],");
}
src.push_str("];");
src.push_str("for test in TESTS {");
for (i, arg) in retptr.iter().enumerate() {
src.push_str(&format!("let mut argret{} = {};", i, arg.default()));
}
src.push_str("let output = ");
src.push_str(&function.name);
src.push_str("(");
for (i, arg) in function.args.iter().enumerate() {
src.push_str(&match arg {
Ty::F32 => format!("f32::from_bits(test[{}] as u32)", i),
Ty::F64 => format!("f64::from_bits(test[{}] as u64)", i),
Ty::I32 => format!("test[{}] as i32", i),
Ty::Bool => format!("test[{}] as i32", i),
});
src.push_str(",");
}
for (i, _) in retptr.iter().enumerate() {
src.push_str(&format!("&mut argret{},", i));
}
src.push_str(");");
if let Some(ty) = &ret {
src.push_str(&format!("let output = output{};", ty.to_i64()));
src.push_str("result.extend_from_slice(&output.to_le_bytes());");
}
for (i, ret) in retptr.iter().enumerate() {
src.push_str(&format!(
"result.extend_from_slice(&(argret{}{}).to_le_bytes());",
i,
ret.to_i64(),
));
}
src.push_str("}");
src.push_str("}");
}
src.push_str("std::io::stdout().write_all(&result).unwrap();");
src.push_str("}");
let path = format!("{}/gen.rs", dst);
fs::write(&path, src).unwrap();
// Make it somewhat pretty if something goes wrong
drop(Command::new("rustfmt").arg(&path).status());
// Compile and execute this tests for the musl target, assuming we're an
// x86_64 host effectively.
let status = Command::new("rustc")
.current_dir(&dst)
.arg(&path)
.arg("--target=x86_64-unknown-linux-musl")
.status()
.unwrap();
assert!(status.success());
let output = Command::new("./gen").current_dir(&dst).output().unwrap();
assert!(output.status.success());
assert!(output.stderr.is_empty());
// Map all the output bytes back to an `i64` and then shove it all into
// the expected results.
let mut results = output.stdout.chunks_exact(8).map(|buf| {
let mut exact = [0; 8];
exact.copy_from_slice(buf);
i64::from_le_bytes(exact)
});
for f in functions.iter_mut() {
for test in f.tests.iter_mut() {
test.outputs = (0..f.ret.len()).map(|_| results.next().unwrap()).collect();
}
}
assert!(results.next().is_none());
}
/// Codegens a file which has a ton of `#[test]` annotations for all the
/// tests that we generated above.
fn generate_unit_tests(functions: &[Function]) {
let mut src = String::new();
let dst = std::env::var("OUT_DIR").unwrap();
for function in functions {
src.push_str("#[test]");
src.push_str("fn ");
src.push_str(&function.name);
src.push_str("_matches_musl() {");
src.push_str(&format!(
"static TESTS: &[([i64; {}], [i64; {}])]",
function.args.len(),
function.ret.len(),
));
src.push_str(" = &[");
for test in function.tests.iter() {
src.push_str("([");
for val in test.inputs.iter() {
src.push_str(&val.to_string());
src.push_str(",");
}
src.push_str("],");
src.push_str("[");
for val in test.outputs.iter() {
src.push_str(&val.to_string());
src.push_str(",");
}
src.push_str("],");
src.push_str("),");
}
src.push_str("];");
src.push_str("for (test, expected) in TESTS {");
src.push_str("let output = ");
src.push_str(&function.name);
src.push_str("(");
for (i, arg) in function.args.iter().enumerate() {
src.push_str(&match arg {
Ty::F32 => format!("f32::from_bits(test[{}] as u32)", i),
Ty::F64 => format!("f64::from_bits(test[{}] as u64)", i),
Ty::I32 => format!("test[{}] as i32", i),
Ty::Bool => format!("test[{}] as i32", i),
});
src.push_str(",");
}
src.push_str(");");
for (i, ret) in function.ret.iter().enumerate() {
let get = if function.ret.len() == 1 { String::new() } else { format!(".{}", i) };
src.push_str(&(match ret {
Ty::F32 => format!("if _eqf(output{}, f32::from_bits(expected[{}] as u32)).is_ok() {{ continue }}", get, i),
Ty::F64 => format!("if _eq(output{}, f64::from_bits(expected[{}] as u64)).is_ok() {{ continue }}", get, i),
Ty::I32 => format!("if output{} as i64 == expected[{}] {{ continue }}", get, i),
Ty::Bool => unreachable!(),
}));
}
src.push_str(
r#"
panic!("INPUT: {:?} EXPECTED: {:?} ACTUAL {:?}", test, expected, output);
"#,
);
src.push_str("}");
src.push_str("}");
}
let path = format!("{}/musl-tests.rs", dst);
fs::write(&path, src).unwrap();
// Try to make it somewhat pretty
drop(Command::new("rustfmt").arg(&path).status());
}
}

View File

@ -0,0 +1,51 @@
//! libm in pure Rust
#![deny(warnings)]
#![no_std]
#![cfg_attr(all(feature = "unstable"), feature(core_intrinsics))]
#![allow(clippy::unreadable_literal)]
#![allow(clippy::many_single_char_names)]
#![allow(clippy::needless_return)]
#![allow(clippy::int_plus_one)]
#![allow(clippy::deprecated_cfg_attr)]
#![allow(clippy::mixed_case_hex_literals)]
#![allow(clippy::float_cmp)]
#![allow(clippy::eq_op)]
#![allow(clippy::assign_op_pattern)]
mod libm_helper;
mod math;
use core::{f32, f64};
pub use self::math::*;
pub use libm_helper::*;
/// Approximate equality with 1 ULP of tolerance
#[doc(hidden)]
#[inline]
pub fn _eqf(a: f32, b: f32) -> Result<(), u32> {
if a.is_nan() && b.is_nan() {
Ok(())
} else {
let err = (a.to_bits() as i32).wrapping_sub(b.to_bits() as i32).abs();
if err <= 1 { Ok(()) } else { Err(err as u32) }
}
}
#[doc(hidden)]
#[inline]
pub fn _eq(a: f64, b: f64) -> Result<(), u64> {
if a.is_nan() && b.is_nan() {
Ok(())
} else {
let err = (a.to_bits() as i64).wrapping_sub(b.to_bits() as i64).abs();
if err <= 1 { Ok(()) } else { Err(err as u64) }
}
}
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
#[cfg(not(target_arch = "powerpc64"))]
#[cfg(all(test, feature = "musl-reference-tests"))]
include!(concat!(env!("OUT_DIR"), "/musl-tests.rs"));

View File

@ -0,0 +1,171 @@
use core::marker::PhantomData;
use crate::*;
/// Generic helper for libm functions, abstracting over f32 and f64. <br/>
/// # Type Parameter:
/// - `T`: Either `f32` or `f64`
///
/// # Examples
/// ```rust
/// use libm::{self, Libm};
///
/// const PI_F32: f32 = 3.1415927410e+00;
/// const PI_F64: f64 = 3.1415926535897931160e+00;
///
/// assert!(Libm::<f32>::cos(0.0f32) == libm::cosf(0.0));
/// assert!(Libm::<f32>::sin(PI_F32) == libm::sinf(PI_F32));
///
/// assert!(Libm::<f64>::cos(0.0f64) == libm::cos(0.0));
/// assert!(Libm::<f64>::sin(PI_F64) == libm::sin(PI_F64));
/// ```
pub struct Libm<T>(PhantomData<T>);
macro_rules! libm_helper {
($t:ident, funcs: $funcs:tt) => {
impl Libm<$t> {
#![allow(unused_parens)]
libm_helper! { $funcs }
}
};
({$($func:tt);*}) => {
$(
libm_helper! { $func }
)*
};
((fn $func:ident($($arg:ident: $arg_typ:ty),*) -> ($($ret_typ:ty),*); => $libm_fn:ident)) => {
#[inline(always)]
pub fn $func($($arg: $arg_typ),*) -> ($($ret_typ),*) {
$libm_fn($($arg),*)
}
};
}
libm_helper! {
f32,
funcs: {
(fn acos(x: f32) -> (f32); => acosf);
(fn acosh(x: f32) -> (f32); => acoshf);
(fn asin(x: f32) -> (f32); => asinf);
(fn asinh(x: f32) -> (f32); => asinhf);
(fn atan(x: f32) -> (f32); => atanf);
(fn atan2(y: f32, x: f32) -> (f32); => atan2f);
(fn atanh(x: f32) -> (f32); => atanhf);
(fn cbrt(x: f32) -> (f32); => cbrtf);
(fn ceil(x: f32) -> (f32); => ceilf);
(fn copysign(x: f32, y: f32) -> (f32); => copysignf);
(fn cos(x: f32) -> (f32); => cosf);
(fn cosh(x: f32) -> (f32); => coshf);
(fn erf(x: f32) -> (f32); => erff);
(fn erfc(x: f32) -> (f32); => erfcf);
(fn exp(x: f32) -> (f32); => expf);
(fn exp2(x: f32) -> (f32); => exp2f);
(fn exp10(x: f32) -> (f32); => exp10f);
(fn expm1(x: f32) -> (f32); => expm1f);
(fn fabs(x: f32) -> (f32); => fabsf);
(fn fdim(x: f32, y: f32) -> (f32); => fdimf);
(fn floor(x: f32) -> (f32); => floorf);
(fn fma(x: f32, y: f32, z: f32) -> (f32); => fmaf);
(fn fmax(x: f32, y: f32) -> (f32); => fmaxf);
(fn fmin(x: f32, y: f32) -> (f32); => fminf);
(fn fmod(x: f32, y: f32) -> (f32); => fmodf);
(fn frexp(x: f32) -> (f32, i32); => frexpf);
(fn hypot(x: f32, y: f32) -> (f32); => hypotf);
(fn ilogb(x: f32) -> (i32); => ilogbf);
(fn j0(x: f32) -> (f32); => j0f);
(fn j1(x: f32) -> (f32); => j1f);
(fn jn(n: i32, x: f32) -> (f32); => jnf);
(fn ldexp(x: f32, n: i32) -> (f32); => ldexpf);
(fn lgamma_r(x: f32) -> (f32, i32); => lgammaf_r);
(fn lgamma(x: f32) -> (f32); => lgammaf);
(fn log(x: f32) -> (f32); => logf);
(fn log1p(x: f32) -> (f32); => log1pf);
(fn log2(x: f32) -> (f32); => log2f);
(fn log10(x: f32) -> (f32); => log10f);
(fn modf(x: f32) -> (f32, f32); => modff);
(fn nextafter(x: f32, y: f32) -> (f32); => nextafterf);
(fn pow(x: f32, y: f32) -> (f32); => powf);
(fn remainder(x: f32, y: f32) -> (f32); => remainderf);
(fn remquo(x: f32, y: f32) -> (f32, i32); => remquof);
(fn rint(x: f32) -> (f32); => rintf);
(fn round(x: f32) -> (f32); => roundf);
(fn scalbn(x: f32, n: i32) -> (f32); => scalbnf);
(fn sin(x: f32) -> (f32); => sinf);
(fn sincos(x: f32) -> (f32, f32); => sincosf);
(fn sinh(x: f32) -> (f32); => sinhf);
(fn sqrt(x: f32) -> (f32); => sqrtf);
(fn tan(x: f32) -> (f32); => tanf);
(fn tanh(x: f32) -> (f32); => tanhf);
(fn tgamma(x: f32) -> (f32); => tgammaf);
(fn trunc(x: f32) -> (f32); => truncf);
(fn y0(x: f32) -> (f32); => y0f);
(fn y1(x: f32) -> (f32); => y1f);
(fn yn(n: i32, x: f32) -> (f32); => ynf)
}
}
libm_helper! {
f64,
funcs: {
(fn acos(x: f64) -> (f64); => acos);
(fn acosh(x: f64) -> (f64); => acosh);
(fn asin(x: f64) -> (f64); => asin);
(fn asinh(x: f64) -> (f64); => asinh);
(fn atan(x: f64) -> (f64); => atan);
(fn atan2(y: f64, x: f64) -> (f64); => atan2);
(fn atanh(x: f64) -> (f64); => atanh);
(fn cbrt(x: f64) -> (f64); => cbrt);
(fn ceil(x: f64) -> (f64); => ceil);
(fn copysign(x: f64, y: f64) -> (f64); => copysign);
(fn cos(x: f64) -> (f64); => cos);
(fn cosh(x: f64) -> (f64); => cosh);
(fn erf(x: f64) -> (f64); => erf);
(fn erfc(x: f64) -> (f64); => erfc);
(fn exp(x: f64) -> (f64); => exp);
(fn exp2(x: f64) -> (f64); => exp2);
(fn exp10(x: f64) -> (f64); => exp10);
(fn expm1(x: f64) -> (f64); => expm1);
(fn fabs(x: f64) -> (f64); => fabs);
(fn fdim(x: f64, y: f64) -> (f64); => fdim);
(fn floor(x: f64) -> (f64); => floor);
(fn fma(x: f64, y: f64, z: f64) -> (f64); => fma);
(fn fmax(x: f64, y: f64) -> (f64); => fmax);
(fn fmin(x: f64, y: f64) -> (f64); => fmin);
(fn fmod(x: f64, y: f64) -> (f64); => fmod);
(fn frexp(x: f64) -> (f64, i32); => frexp);
(fn hypot(x: f64, y: f64) -> (f64); => hypot);
(fn ilogb(x: f64) -> (i32); => ilogb);
(fn j0(x: f64) -> (f64); => j0);
(fn j1(x: f64) -> (f64); => j1);
(fn jn(n: i32, x: f64) -> (f64); => jn);
(fn ldexp(x: f64, n: i32) -> (f64); => ldexp);
(fn lgamma_r(x: f64) -> (f64, i32); => lgamma_r);
(fn lgamma(x: f64) -> (f64); => lgamma);
(fn log(x: f64) -> (f64); => log);
(fn log1p(x: f64) -> (f64); => log1p);
(fn log2(x: f64) -> (f64); => log2);
(fn log10(x: f64) -> (f64); => log10);
(fn modf(x: f64) -> (f64, f64); => modf);
(fn nextafter(x: f64, y: f64) -> (f64); => nextafter);
(fn pow(x: f64, y: f64) -> (f64); => pow);
(fn remainder(x: f64, y: f64) -> (f64); => remainder);
(fn remquo(x: f64, y: f64) -> (f64, i32); => remquo);
(fn rint(x: f64) -> (f64); => rint);
(fn round(x: f64) -> (f64); => round);
(fn scalbn(x: f64, n: i32) -> (f64); => scalbn);
(fn sin(x: f64) -> (f64); => sin);
(fn sincos(x: f64) -> (f64, f64); => sincos);
(fn sinh(x: f64) -> (f64); => sinh);
(fn sqrt(x: f64) -> (f64); => sqrt);
(fn tan(x: f64) -> (f64); => tan);
(fn tanh(x: f64) -> (f64); => tanh);
(fn tgamma(x: f64) -> (f64); => tgamma);
(fn trunc(x: f64) -> (f64); => trunc);
(fn y0(x: f64) -> (f64); => y0);
(fn y1(x: f64) -> (f64); => y1);
(fn yn(n: i32, x: f64) -> (f64); => yn)
}
}

View File

@ -0,0 +1,112 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_acos.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* acos(x)
* Method :
* acos(x) = pi/2 - asin(x)
* acos(-x) = pi/2 + asin(x)
* For |x|<=0.5
* acos(x) = pi/2 - (x + x*x^2*R(x^2)) (see asin.c)
* For x>0.5
* acos(x) = pi/2 - (pi/2 - 2asin(sqrt((1-x)/2)))
* = 2asin(sqrt((1-x)/2))
* = 2s + 2s*z*R(z) ...z=(1-x)/2, s=sqrt(z)
* = 2f + (2c + 2s*z*R(z))
* where f=hi part of s, and c = (z-f*f)/(s+f) is the correction term
* for f so that f+c ~ sqrt(z).
* For x<-0.5
* acos(x) = pi - 2asin(sqrt((1-|x|)/2))
* = pi - 0.5*(s+s*z*R(z)), where z=(1-|x|)/2,s=sqrt(z)
*
* Special cases:
* if x is NaN, return x itself;
* if |x|>1, return NaN with invalid signal.
*
* Function needed: sqrt
*/
use super::sqrt;
const PIO2_HI: f64 = 1.57079632679489655800e+00; /* 0x3FF921FB, 0x54442D18 */
const PIO2_LO: f64 = 6.12323399573676603587e-17; /* 0x3C91A626, 0x33145C07 */
const PS0: f64 = 1.66666666666666657415e-01; /* 0x3FC55555, 0x55555555 */
const PS1: f64 = -3.25565818622400915405e-01; /* 0xBFD4D612, 0x03EB6F7D */
const PS2: f64 = 2.01212532134862925881e-01; /* 0x3FC9C155, 0x0E884455 */
const PS3: f64 = -4.00555345006794114027e-02; /* 0xBFA48228, 0xB5688F3B */
const PS4: f64 = 7.91534994289814532176e-04; /* 0x3F49EFE0, 0x7501B288 */
const PS5: f64 = 3.47933107596021167570e-05; /* 0x3F023DE1, 0x0DFDF709 */
const QS1: f64 = -2.40339491173441421878e+00; /* 0xC0033A27, 0x1C8A2D4B */
const QS2: f64 = 2.02094576023350569471e+00; /* 0x40002AE5, 0x9C598AC8 */
const QS3: f64 = -6.88283971605453293030e-01; /* 0xBFE6066C, 0x1B8D0159 */
const QS4: f64 = 7.70381505559019352791e-02; /* 0x3FB3B8C5, 0xB12E9282 */
fn r(z: f64) -> f64 {
let p: f64 = z * (PS0 + z * (PS1 + z * (PS2 + z * (PS3 + z * (PS4 + z * PS5)))));
let q: f64 = 1.0 + z * (QS1 + z * (QS2 + z * (QS3 + z * QS4)));
p / q
}
/// Arccosine (f64)
///
/// Computes the inverse cosine (arc cosine) of the input value.
/// Arguments must be in the range -1 to 1.
/// Returns values in radians, in the range of 0 to pi.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn acos(x: f64) -> f64 {
let x1p_120f = f64::from_bits(0x3870000000000000); // 0x1p-120 === 2 ^ -120
let z: f64;
let w: f64;
let s: f64;
let c: f64;
let df: f64;
let hx: u32;
let ix: u32;
hx = (x.to_bits() >> 32) as u32;
ix = hx & 0x7fffffff;
/* |x| >= 1 or nan */
if ix >= 0x3ff00000 {
let lx: u32 = x.to_bits() as u32;
if ((ix - 0x3ff00000) | lx) == 0 {
/* acos(1)=0, acos(-1)=pi */
if (hx >> 31) != 0 {
return 2. * PIO2_HI + x1p_120f;
}
return 0.;
}
return 0. / (x - x);
}
/* |x| < 0.5 */
if ix < 0x3fe00000 {
if ix <= 0x3c600000 {
/* |x| < 2**-57 */
return PIO2_HI + x1p_120f;
}
return PIO2_HI - (x - (PIO2_LO - x * r(x * x)));
}
/* x < -0.5 */
if (hx >> 31) != 0 {
z = (1.0 + x) * 0.5;
s = sqrt(z);
w = r(z) * s - PIO2_LO;
return 2. * (PIO2_HI - (s + w));
}
/* x > 0.5 */
z = (1.0 - x) * 0.5;
s = sqrt(z);
// Set the low 4 bytes to zero
df = f64::from_bits(s.to_bits() & 0xff_ff_ff_ff_00_00_00_00);
c = (z - df * df) / (s + df);
w = r(z) * s + c;
2. * (df + w)
}

View File

@ -0,0 +1,79 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_acosf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::sqrtf::sqrtf;
const PIO2_HI: f32 = 1.5707962513e+00; /* 0x3fc90fda */
const PIO2_LO: f32 = 7.5497894159e-08; /* 0x33a22168 */
const P_S0: f32 = 1.6666586697e-01;
const P_S1: f32 = -4.2743422091e-02;
const P_S2: f32 = -8.6563630030e-03;
const Q_S1: f32 = -7.0662963390e-01;
fn r(z: f32) -> f32 {
let p = z * (P_S0 + z * (P_S1 + z * P_S2));
let q = 1. + z * Q_S1;
p / q
}
/// Arccosine (f32)
///
/// Computes the inverse cosine (arc cosine) of the input value.
/// Arguments must be in the range -1 to 1.
/// Returns values in radians, in the range of 0 to pi.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn acosf(x: f32) -> f32 {
let x1p_120 = f32::from_bits(0x03800000); // 0x1p-120 === 2 ^ (-120)
let z: f32;
let w: f32;
let s: f32;
let mut hx = x.to_bits();
let ix = hx & 0x7fffffff;
/* |x| >= 1 or nan */
if ix >= 0x3f800000 {
if ix == 0x3f800000 {
if (hx >> 31) != 0 {
return 2. * PIO2_HI + x1p_120;
}
return 0.;
}
return 0. / (x - x);
}
/* |x| < 0.5 */
if ix < 0x3f000000 {
if ix <= 0x32800000 {
/* |x| < 2**-26 */
return PIO2_HI + x1p_120;
}
return PIO2_HI - (x - (PIO2_LO - x * r(x * x)));
}
/* x < -0.5 */
if (hx >> 31) != 0 {
z = (1. + x) * 0.5;
s = sqrtf(z);
w = r(z) * s - PIO2_LO;
return 2. * (PIO2_HI - (s + w));
}
/* x > 0.5 */
z = (1. - x) * 0.5;
s = sqrtf(z);
hx = s.to_bits();
let df = f32::from_bits(hx & 0xfffff000);
let c = (z - df * df) / (s + df);
w = r(z) * s + c;
2. * (df + w)
}

View File

@ -0,0 +1,27 @@
use super::{log, log1p, sqrt};
const LN2: f64 = 0.693147180559945309417232121458176568; /* 0x3fe62e42, 0xfefa39ef*/
/// Inverse hyperbolic cosine (f64)
///
/// Calculates the inverse hyperbolic cosine of `x`.
/// Is defined as `log(x + sqrt(x*x-1))`.
/// `x` must be a number greater than or equal to 1.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn acosh(x: f64) -> f64 {
let u = x.to_bits();
let e = ((u >> 52) as usize) & 0x7ff;
/* x < 1 domain error is handled in the called functions */
if e < 0x3ff + 1 {
/* |x| < 2, up to 2ulp error in [1,1.125] */
return log1p(x - 1.0 + sqrt((x - 1.0) * (x - 1.0) + 2.0 * (x - 1.0)));
}
if e < 0x3ff + 26 {
/* |x| < 0x1p26 */
return log(2.0 * x - 1.0 / (x + sqrt(x * x - 1.0)));
}
/* |x| >= 0x1p26 or nan */
return log(x) + LN2;
}

View File

@ -0,0 +1,26 @@
use super::{log1pf, logf, sqrtf};
const LN2: f32 = 0.693147180559945309417232121458176568;
/// Inverse hyperbolic cosine (f32)
///
/// Calculates the inverse hyperbolic cosine of `x`.
/// Is defined as `log(x + sqrt(x*x-1))`.
/// `x` must be a number greater than or equal to 1.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn acoshf(x: f32) -> f32 {
let u = x.to_bits();
let a = u & 0x7fffffff;
if a < 0x3f800000 + (1 << 23) {
/* |x| < 2, invalid if x < 1 or nan */
/* up to 2ulp error in [1,1.125] */
return log1pf(x - 1.0 + sqrtf((x - 1.0) * (x - 1.0) + 2.0 * (x - 1.0)));
}
if a < 0x3f800000 + (12 << 23) {
/* |x| < 0x1p12 */
return logf(2.0 * x - 1.0 / (x + sqrtf(x * x - 1.0)));
}
/* x >= 0x1p12 */
return logf(x) + LN2;
}

View File

@ -0,0 +1,115 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_asin.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* asin(x)
* Method :
* Since asin(x) = x + x^3/6 + x^5*3/40 + x^7*15/336 + ...
* we approximate asin(x) on [0,0.5] by
* asin(x) = x + x*x^2*R(x^2)
* where
* R(x^2) is a rational approximation of (asin(x)-x)/x^3
* and its remez error is bounded by
* |(asin(x)-x)/x^3 - R(x^2)| < 2^(-58.75)
*
* For x in [0.5,1]
* asin(x) = pi/2-2*asin(sqrt((1-x)/2))
* Let y = (1-x), z = y/2, s := sqrt(z), and pio2_hi+pio2_lo=pi/2;
* then for x>0.98
* asin(x) = pi/2 - 2*(s+s*z*R(z))
* = pio2_hi - (2*(s+s*z*R(z)) - pio2_lo)
* For x<=0.98, let pio4_hi = pio2_hi/2, then
* f = hi part of s;
* c = sqrt(z) - f = (z-f*f)/(s+f) ...f+c=sqrt(z)
* and
* asin(x) = pi/2 - 2*(s+s*z*R(z))
* = pio4_hi+(pio4-2s)-(2s*z*R(z)-pio2_lo)
* = pio4_hi+(pio4-2f)-(2s*z*R(z)-(pio2_lo+2c))
*
* Special cases:
* if x is NaN, return x itself;
* if |x|>1, return NaN with invalid signal.
*
*/
use super::{fabs, get_high_word, get_low_word, sqrt, with_set_low_word};
const PIO2_HI: f64 = 1.57079632679489655800e+00; /* 0x3FF921FB, 0x54442D18 */
const PIO2_LO: f64 = 6.12323399573676603587e-17; /* 0x3C91A626, 0x33145C07 */
/* coefficients for R(x^2) */
const P_S0: f64 = 1.66666666666666657415e-01; /* 0x3FC55555, 0x55555555 */
const P_S1: f64 = -3.25565818622400915405e-01; /* 0xBFD4D612, 0x03EB6F7D */
const P_S2: f64 = 2.01212532134862925881e-01; /* 0x3FC9C155, 0x0E884455 */
const P_S3: f64 = -4.00555345006794114027e-02; /* 0xBFA48228, 0xB5688F3B */
const P_S4: f64 = 7.91534994289814532176e-04; /* 0x3F49EFE0, 0x7501B288 */
const P_S5: f64 = 3.47933107596021167570e-05; /* 0x3F023DE1, 0x0DFDF709 */
const Q_S1: f64 = -2.40339491173441421878e+00; /* 0xC0033A27, 0x1C8A2D4B */
const Q_S2: f64 = 2.02094576023350569471e+00; /* 0x40002AE5, 0x9C598AC8 */
const Q_S3: f64 = -6.88283971605453293030e-01; /* 0xBFE6066C, 0x1B8D0159 */
const Q_S4: f64 = 7.70381505559019352791e-02; /* 0x3FB3B8C5, 0xB12E9282 */
fn comp_r(z: f64) -> f64 {
let p = z * (P_S0 + z * (P_S1 + z * (P_S2 + z * (P_S3 + z * (P_S4 + z * P_S5)))));
let q = 1.0 + z * (Q_S1 + z * (Q_S2 + z * (Q_S3 + z * Q_S4)));
p / q
}
/// Arcsine (f64)
///
/// Computes the inverse sine (arc sine) of the argument `x`.
/// Arguments to asin must be in the range -1 to 1.
/// Returns values in radians, in the range of -pi/2 to pi/2.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn asin(mut x: f64) -> f64 {
let z: f64;
let r: f64;
let s: f64;
let hx: u32;
let ix: u32;
hx = get_high_word(x);
ix = hx & 0x7fffffff;
/* |x| >= 1 or nan */
if ix >= 0x3ff00000 {
let lx: u32;
lx = get_low_word(x);
if ((ix - 0x3ff00000) | lx) == 0 {
/* asin(1) = +-pi/2 with inexact */
return x * PIO2_HI + f64::from_bits(0x3870000000000000);
} else {
return 0.0 / (x - x);
}
}
/* |x| < 0.5 */
if ix < 0x3fe00000 {
/* if 0x1p-1022 <= |x| < 0x1p-26, avoid raising underflow */
if ix < 0x3e500000 && ix >= 0x00100000 {
return x;
} else {
return x + x * comp_r(x * x);
}
}
/* 1 > |x| >= 0.5 */
z = (1.0 - fabs(x)) * 0.5;
s = sqrt(z);
r = comp_r(z);
if ix >= 0x3fef3333 {
/* if |x| > 0.975 */
x = PIO2_HI - (2. * (s + s * r) - PIO2_LO);
} else {
let f: f64;
let c: f64;
/* f+c = sqrt(z) */
f = with_set_low_word(s, 0);
c = (z - f * f) / (s + f);
x = 0.5 * PIO2_HI - (2.0 * s * r - (PIO2_LO - 2.0 * c) - (0.5 * PIO2_HI - 2.0 * f));
}
if hx >> 31 != 0 { -x } else { x }
}

View File

@ -0,0 +1,68 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_asinf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::fabsf::fabsf;
use super::sqrt::sqrt;
const PIO2: f64 = 1.570796326794896558e+00;
/* coefficients for R(x^2) */
const P_S0: f32 = 1.6666586697e-01;
const P_S1: f32 = -4.2743422091e-02;
const P_S2: f32 = -8.6563630030e-03;
const Q_S1: f32 = -7.0662963390e-01;
fn r(z: f32) -> f32 {
let p = z * (P_S0 + z * (P_S1 + z * P_S2));
let q = 1. + z * Q_S1;
p / q
}
/// Arcsine (f32)
///
/// Computes the inverse sine (arc sine) of the argument `x`.
/// Arguments to asin must be in the range -1 to 1.
/// Returns values in radians, in the range of -pi/2 to pi/2.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn asinf(mut x: f32) -> f32 {
let x1p_120 = f64::from_bits(0x3870000000000000); // 0x1p-120 === 2 ^ (-120)
let hx = x.to_bits();
let ix = hx & 0x7fffffff;
if ix >= 0x3f800000 {
/* |x| >= 1 */
if ix == 0x3f800000 {
/* |x| == 1 */
return ((x as f64) * PIO2 + x1p_120) as f32; /* asin(+-1) = +-pi/2 with inexact */
}
return 0. / (x - x); /* asin(|x|>1) is NaN */
}
if ix < 0x3f000000 {
/* |x| < 0.5 */
/* if 0x1p-126 <= |x| < 0x1p-12, avoid raising underflow */
if (ix < 0x39800000) && (ix >= 0x00800000) {
return x;
}
return x + x * r(x * x);
}
/* 1 > |x| >= 0.5 */
let z = (1. - fabsf(x)) * 0.5;
let s = sqrt(z as f64);
x = (PIO2 - 2. * (s + s * (r(z) as f64))) as f32;
if (hx >> 31) != 0 { -x } else { x }
}

View File

@ -0,0 +1,36 @@
use super::{log, log1p, sqrt};
const LN2: f64 = 0.693147180559945309417232121458176568; /* 0x3fe62e42, 0xfefa39ef*/
/* asinh(x) = sign(x)*log(|x|+sqrt(x*x+1)) ~= x - x^3/6 + o(x^5) */
/// Inverse hyperbolic sine (f64)
///
/// Calculates the inverse hyperbolic sine of `x`.
/// Is defined as `sgn(x)*log(|x|+sqrt(x*x+1))`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn asinh(mut x: f64) -> f64 {
let mut u = x.to_bits();
let e = ((u >> 52) as usize) & 0x7ff;
let sign = (u >> 63) != 0;
/* |x| */
u &= (!0) >> 1;
x = f64::from_bits(u);
if e >= 0x3ff + 26 {
/* |x| >= 0x1p26 or inf or nan */
x = log(x) + LN2;
} else if e >= 0x3ff + 1 {
/* |x| >= 2 */
x = log(2.0 * x + 1.0 / (sqrt(x * x + 1.0) + x));
} else if e >= 0x3ff - 26 {
/* |x| >= 0x1p-26, up to 1.6ulp error in [0.125,0.5] */
x = log1p(x + x * x / (sqrt(x * x + 1.0) + 1.0));
} else {
/* |x| < 0x1p-26, raise inexact if x != 0 */
let x1p120 = f64::from_bits(0x4770000000000000);
force_eval!(x + x1p120);
}
if sign { -x } else { x }
}

View File

@ -0,0 +1,35 @@
use super::{log1pf, logf, sqrtf};
const LN2: f32 = 0.693147180559945309417232121458176568;
/* asinh(x) = sign(x)*log(|x|+sqrt(x*x+1)) ~= x - x^3/6 + o(x^5) */
/// Inverse hyperbolic sine (f32)
///
/// Calculates the inverse hyperbolic sine of `x`.
/// Is defined as `sgn(x)*log(|x|+sqrt(x*x+1))`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn asinhf(mut x: f32) -> f32 {
let u = x.to_bits();
let i = u & 0x7fffffff;
let sign = (u >> 31) != 0;
/* |x| */
x = f32::from_bits(i);
if i >= 0x3f800000 + (12 << 23) {
/* |x| >= 0x1p12 or inf or nan */
x = logf(x) + LN2;
} else if i >= 0x3f800000 + (1 << 23) {
/* |x| >= 2 */
x = logf(2.0 * x + 1.0 / (sqrtf(x * x + 1.0) + x));
} else if i >= 0x3f800000 - (12 << 23) {
/* |x| >= 0x1p-12, up to 1.6ulp error in [0.125,0.5] */
x = log1pf(x + x * x / (sqrtf(x * x + 1.0) + 1.0));
} else {
/* |x| < 0x1p-12, raise inexact if x!=0 */
let x1p120 = f32::from_bits(0x7b800000);
force_eval!(x + x1p120);
}
if sign { -x } else { x }
}

View File

@ -0,0 +1,180 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_atan.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* atan(x)
* Method
* 1. Reduce x to positive by atan(x) = -atan(-x).
* 2. According to the integer k=4t+0.25 chopped, t=x, the argument
* is further reduced to one of the following intervals and the
* arctangent of t is evaluated by the corresponding formula:
*
* [0,7/16] atan(x) = t-t^3*(a1+t^2*(a2+...(a10+t^2*a11)...)
* [7/16,11/16] atan(x) = atan(1/2) + atan( (t-0.5)/(1+t/2) )
* [11/16.19/16] atan(x) = atan( 1 ) + atan( (t-1)/(1+t) )
* [19/16,39/16] atan(x) = atan(3/2) + atan( (t-1.5)/(1+1.5t) )
* [39/16,INF] atan(x) = atan(INF) + atan( -1/t )
*
* Constants:
* The hexadecimal values are the intended ones for the following
* constants. The decimal values may be used, provided that the
* compiler will convert from decimal to binary accurately enough
* to produce the hexadecimal values shown.
*/
use super::fabs;
use core::f64;
const ATANHI: [f64; 4] = [
4.63647609000806093515e-01, /* atan(0.5)hi 0x3FDDAC67, 0x0561BB4F */
7.85398163397448278999e-01, /* atan(1.0)hi 0x3FE921FB, 0x54442D18 */
9.82793723247329054082e-01, /* atan(1.5)hi 0x3FEF730B, 0xD281F69B */
1.57079632679489655800e+00, /* atan(inf)hi 0x3FF921FB, 0x54442D18 */
];
const ATANLO: [f64; 4] = [
2.26987774529616870924e-17, /* atan(0.5)lo 0x3C7A2B7F, 0x222F65E2 */
3.06161699786838301793e-17, /* atan(1.0)lo 0x3C81A626, 0x33145C07 */
1.39033110312309984516e-17, /* atan(1.5)lo 0x3C700788, 0x7AF0CBBD */
6.12323399573676603587e-17, /* atan(inf)lo 0x3C91A626, 0x33145C07 */
];
const AT: [f64; 11] = [
3.33333333333329318027e-01, /* 0x3FD55555, 0x5555550D */
-1.99999999998764832476e-01, /* 0xBFC99999, 0x9998EBC4 */
1.42857142725034663711e-01, /* 0x3FC24924, 0x920083FF */
-1.11111104054623557880e-01, /* 0xBFBC71C6, 0xFE231671 */
9.09088713343650656196e-02, /* 0x3FB745CD, 0xC54C206E */
-7.69187620504482999495e-02, /* 0xBFB3B0F2, 0xAF749A6D */
6.66107313738753120669e-02, /* 0x3FB10D66, 0xA0D03D51 */
-5.83357013379057348645e-02, /* 0xBFADDE2D, 0x52DEFD9A */
4.97687799461593236017e-02, /* 0x3FA97B4B, 0x24760DEB */
-3.65315727442169155270e-02, /* 0xBFA2B444, 0x2C6A6C2F */
1.62858201153657823623e-02, /* 0x3F90AD3A, 0xE322DA11 */
];
/// Arctangent (f64)
///
/// Computes the inverse tangent (arc tangent) of the input value.
/// Returns a value in radians, in the range of -pi/2 to pi/2.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atan(x: f64) -> f64 {
let mut x = x;
let mut ix = (x.to_bits() >> 32) as u32;
let sign = ix >> 31;
ix &= 0x7fff_ffff;
if ix >= 0x4410_0000 {
if x.is_nan() {
return x;
}
let z = ATANHI[3] + f64::from_bits(0x0380_0000); // 0x1p-120f
return if sign != 0 { -z } else { z };
}
let id = if ix < 0x3fdc_0000 {
/* |x| < 0.4375 */
if ix < 0x3e40_0000 {
/* |x| < 2^-27 */
if ix < 0x0010_0000 {
/* raise underflow for subnormal x */
force_eval!(x as f32);
}
return x;
}
-1
} else {
x = fabs(x);
if ix < 0x3ff30000 {
/* |x| < 1.1875 */
if ix < 0x3fe60000 {
/* 7/16 <= |x| < 11/16 */
x = (2. * x - 1.) / (2. + x);
0
} else {
/* 11/16 <= |x| < 19/16 */
x = (x - 1.) / (x + 1.);
1
}
} else if ix < 0x40038000 {
/* |x| < 2.4375 */
x = (x - 1.5) / (1. + 1.5 * x);
2
} else {
/* 2.4375 <= |x| < 2^66 */
x = -1. / x;
3
}
};
let z = x * x;
let w = z * z;
/* break sum from i=0 to 10 AT[i]z**(i+1) into odd and even poly */
let s1 = z * (AT[0] + w * (AT[2] + w * (AT[4] + w * (AT[6] + w * (AT[8] + w * AT[10])))));
let s2 = w * (AT[1] + w * (AT[3] + w * (AT[5] + w * (AT[7] + w * AT[9]))));
if id < 0 {
return x - x * (s1 + s2);
}
let z = i!(ATANHI, id as usize) - (x * (s1 + s2) - i!(ATANLO, id as usize) - x);
if sign != 0 { -z } else { z }
}
#[cfg(test)]
mod tests {
use super::atan;
use core::f64;
#[test]
fn sanity_check() {
for (input, answer) in [
(3.0_f64.sqrt() / 3.0, f64::consts::FRAC_PI_6),
(1.0, f64::consts::FRAC_PI_4),
(3.0_f64.sqrt(), f64::consts::FRAC_PI_3),
(-3.0_f64.sqrt() / 3.0, -f64::consts::FRAC_PI_6),
(-1.0, -f64::consts::FRAC_PI_4),
(-3.0_f64.sqrt(), -f64::consts::FRAC_PI_3),
]
.iter()
{
assert!(
(atan(*input) - answer) / answer < 1e-5,
"\natan({:.4}/16) = {:.4}, actual: {}",
input * 16.0,
answer,
atan(*input)
);
}
}
#[test]
fn zero() {
assert_eq!(atan(0.0), 0.0);
}
#[test]
fn infinity() {
assert_eq!(atan(f64::INFINITY), f64::consts::FRAC_PI_2);
}
#[test]
fn minus_infinity() {
assert_eq!(atan(f64::NEG_INFINITY), -f64::consts::FRAC_PI_2);
}
#[test]
fn nan() {
assert!(atan(f64::NAN).is_nan());
}
}

View File

@ -0,0 +1,126 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_atan2.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*
*/
/* atan2(y,x)
* Method :
* 1. Reduce y to positive by atan2(y,x)=-atan2(-y,x).
* 2. Reduce x to positive by (if x and y are unexceptional):
* ARG (x+iy) = arctan(y/x) ... if x > 0,
* ARG (x+iy) = pi - arctan[y/(-x)] ... if x < 0,
*
* Special cases:
*
* ATAN2((anything), NaN ) is NaN;
* ATAN2(NAN , (anything) ) is NaN;
* ATAN2(+-0, +(anything but NaN)) is +-0 ;
* ATAN2(+-0, -(anything but NaN)) is +-pi ;
* ATAN2(+-(anything but 0 and NaN), 0) is +-pi/2;
* ATAN2(+-(anything but INF and NaN), +INF) is +-0 ;
* ATAN2(+-(anything but INF and NaN), -INF) is +-pi;
* ATAN2(+-INF,+INF ) is +-pi/4 ;
* ATAN2(+-INF,-INF ) is +-3pi/4;
* ATAN2(+-INF, (anything but,0,NaN, and INF)) is +-pi/2;
*
* Constants:
* The hexadecimal values are the intended ones for the following
* constants. The decimal values may be used, provided that the
* compiler will convert from decimal to binary accurately enough
* to produce the hexadecimal values shown.
*/
use super::atan;
use super::fabs;
const PI: f64 = 3.1415926535897931160E+00; /* 0x400921FB, 0x54442D18 */
const PI_LO: f64 = 1.2246467991473531772E-16; /* 0x3CA1A626, 0x33145C07 */
/// Arctangent of y/x (f64)
///
/// Computes the inverse tangent (arc tangent) of `y/x`.
/// Produces the correct result even for angles near pi/2 or -pi/2 (that is, when `x` is near 0).
/// Returns a value in radians, in the range of -pi to pi.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atan2(y: f64, x: f64) -> f64 {
if x.is_nan() || y.is_nan() {
return x + y;
}
let mut ix = (x.to_bits() >> 32) as u32;
let lx = x.to_bits() as u32;
let mut iy = (y.to_bits() >> 32) as u32;
let ly = y.to_bits() as u32;
if ((ix.wrapping_sub(0x3ff00000)) | lx) == 0 {
/* x = 1.0 */
return atan(y);
}
let m = ((iy >> 31) & 1) | ((ix >> 30) & 2); /* 2*sign(x)+sign(y) */
ix &= 0x7fffffff;
iy &= 0x7fffffff;
/* when y = 0 */
if (iy | ly) == 0 {
return match m {
0 | 1 => y, /* atan(+-0,+anything)=+-0 */
2 => PI, /* atan(+0,-anything) = PI */
_ => -PI, /* atan(-0,-anything) =-PI */
};
}
/* when x = 0 */
if (ix | lx) == 0 {
return if m & 1 != 0 { -PI / 2.0 } else { PI / 2.0 };
}
/* when x is INF */
if ix == 0x7ff00000 {
if iy == 0x7ff00000 {
return match m {
0 => PI / 4.0, /* atan(+INF,+INF) */
1 => -PI / 4.0, /* atan(-INF,+INF) */
2 => 3.0 * PI / 4.0, /* atan(+INF,-INF) */
_ => -3.0 * PI / 4.0, /* atan(-INF,-INF) */
};
} else {
return match m {
0 => 0.0, /* atan(+...,+INF) */
1 => -0.0, /* atan(-...,+INF) */
2 => PI, /* atan(+...,-INF) */
_ => -PI, /* atan(-...,-INF) */
};
}
}
/* |y/x| > 0x1p64 */
if ix.wrapping_add(64 << 20) < iy || iy == 0x7ff00000 {
return if m & 1 != 0 { -PI / 2.0 } else { PI / 2.0 };
}
/* z = atan(|y/x|) without spurious underflow */
let z = if (m & 2 != 0) && iy.wrapping_add(64 << 20) < ix {
/* |y/x| < 0x1p-64, x<0 */
0.0
} else {
atan(fabs(y / x))
};
match m {
0 => z, /* atan(+,+) */
1 => -z, /* atan(-,+) */
2 => PI - (z - PI_LO), /* atan(+,-) */
_ => (z - PI_LO) - PI, /* atan(-,-) */
}
}
#[test]
fn sanity_check() {
assert_eq!(atan2(0.0, 1.0), 0.0);
assert_eq!(atan2(0.0, -1.0), PI);
assert_eq!(atan2(-0.0, -1.0), -PI);
assert_eq!(atan2(3.0, 2.0), atan(3.0 / 2.0));
assert_eq!(atan2(2.0, -1.0), atan(2.0 / -1.0) + PI);
assert_eq!(atan2(-2.0, -1.0), atan(-2.0 / -1.0) - PI);
}

View File

@ -0,0 +1,91 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_atan2f.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::atanf;
use super::fabsf;
const PI: f32 = 3.1415927410e+00; /* 0x40490fdb */
const PI_LO: f32 = -8.7422776573e-08; /* 0xb3bbbd2e */
/// Arctangent of y/x (f32)
///
/// Computes the inverse tangent (arc tangent) of `y/x`.
/// Produces the correct result even for angles near pi/2 or -pi/2 (that is, when `x` is near 0).
/// Returns a value in radians, in the range of -pi to pi.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atan2f(y: f32, x: f32) -> f32 {
if x.is_nan() || y.is_nan() {
return x + y;
}
let mut ix = x.to_bits();
let mut iy = y.to_bits();
if ix == 0x3f800000 {
/* x=1.0 */
return atanf(y);
}
let m = ((iy >> 31) & 1) | ((ix >> 30) & 2); /* 2*sign(x)+sign(y) */
ix &= 0x7fffffff;
iy &= 0x7fffffff;
/* when y = 0 */
if iy == 0 {
return match m {
0 | 1 => y, /* atan(+-0,+anything)=+-0 */
2 => PI, /* atan(+0,-anything) = pi */
3 | _ => -PI, /* atan(-0,-anything) =-pi */
};
}
/* when x = 0 */
if ix == 0 {
return if m & 1 != 0 { -PI / 2. } else { PI / 2. };
}
/* when x is INF */
if ix == 0x7f800000 {
return if iy == 0x7f800000 {
match m {
0 => PI / 4., /* atan(+INF,+INF) */
1 => -PI / 4., /* atan(-INF,+INF) */
2 => 3. * PI / 4., /* atan(+INF,-INF)*/
3 | _ => -3. * PI / 4., /* atan(-INF,-INF)*/
}
} else {
match m {
0 => 0., /* atan(+...,+INF) */
1 => -0., /* atan(-...,+INF) */
2 => PI, /* atan(+...,-INF) */
3 | _ => -PI, /* atan(-...,-INF) */
}
};
}
/* |y/x| > 0x1p26 */
if (ix + (26 << 23) < iy) || (iy == 0x7f800000) {
return if m & 1 != 0 { -PI / 2. } else { PI / 2. };
}
/* z = atan(|y/x|) with correct underflow */
let z = if (m & 2 != 0) && (iy + (26 << 23) < ix) {
/*|y/x| < 0x1p-26, x < 0 */
0.
} else {
atanf(fabsf(y / x))
};
match m {
0 => z, /* atan(+,+) */
1 => -z, /* atan(-,+) */
2 => PI - (z - PI_LO), /* atan(+,-) */
_ => (z - PI_LO) - PI, /* case 3 */ /* atan(-,-) */
}
}

View File

@ -0,0 +1,103 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_atanf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::fabsf;
const ATAN_HI: [f32; 4] = [
4.6364760399e-01, /* atan(0.5)hi 0x3eed6338 */
7.8539812565e-01, /* atan(1.0)hi 0x3f490fda */
9.8279368877e-01, /* atan(1.5)hi 0x3f7b985e */
1.5707962513e+00, /* atan(inf)hi 0x3fc90fda */
];
const ATAN_LO: [f32; 4] = [
5.0121582440e-09, /* atan(0.5)lo 0x31ac3769 */
3.7748947079e-08, /* atan(1.0)lo 0x33222168 */
3.4473217170e-08, /* atan(1.5)lo 0x33140fb4 */
7.5497894159e-08, /* atan(inf)lo 0x33a22168 */
];
const A_T: [f32; 5] =
[3.3333328366e-01, -1.9999158382e-01, 1.4253635705e-01, -1.0648017377e-01, 6.1687607318e-02];
/// Arctangent (f32)
///
/// Computes the inverse tangent (arc tangent) of the input value.
/// Returns a value in radians, in the range of -pi/2 to pi/2.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atanf(mut x: f32) -> f32 {
let x1p_120 = f32::from_bits(0x03800000); // 0x1p-120 === 2 ^ (-120)
let z: f32;
let mut ix = x.to_bits();
let sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix >= 0x4c800000 {
/* if |x| >= 2**26 */
if x.is_nan() {
return x;
}
z = i!(ATAN_HI, 3) + x1p_120;
return if sign { -z } else { z };
}
let id = if ix < 0x3ee00000 {
/* |x| < 0.4375 */
if ix < 0x39800000 {
/* |x| < 2**-12 */
if ix < 0x00800000 {
/* raise underflow for subnormal x */
force_eval!(x * x);
}
return x;
}
-1
} else {
x = fabsf(x);
if ix < 0x3f980000 {
/* |x| < 1.1875 */
if ix < 0x3f300000 {
/* 7/16 <= |x| < 11/16 */
x = (2. * x - 1.) / (2. + x);
0
} else {
/* 11/16 <= |x| < 19/16 */
x = (x - 1.) / (x + 1.);
1
}
} else if ix < 0x401c0000 {
/* |x| < 2.4375 */
x = (x - 1.5) / (1. + 1.5 * x);
2
} else {
/* 2.4375 <= |x| < 2**26 */
x = -1. / x;
3
}
};
/* end of argument reduction */
z = x * x;
let w = z * z;
/* break sum from i=0 to 10 aT[i]z**(i+1) into odd and even poly */
let s1 = z * (i!(A_T, 0) + w * (i!(A_T, 2) + w * i!(A_T, 4)));
let s2 = w * (i!(A_T, 1) + w * i!(A_T, 3));
if id < 0 {
return x - x * (s1 + s2);
}
let id = id as usize;
let z = i!(ATAN_HI, id) - ((x * (s1 + s2) - i!(ATAN_LO, id)) - x);
if sign { -z } else { z }
}

View File

@ -0,0 +1,33 @@
use super::log1p;
/* atanh(x) = log((1+x)/(1-x))/2 = log1p(2x/(1-x))/2 ~= x + x^3/3 + o(x^5) */
/// Inverse hyperbolic tangent (f64)
///
/// Calculates the inverse hyperbolic tangent of `x`.
/// Is defined as `log((1+x)/(1-x))/2 = log1p(2x/(1-x))/2`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atanh(x: f64) -> f64 {
let u = x.to_bits();
let e = ((u >> 52) as usize) & 0x7ff;
let sign = (u >> 63) != 0;
/* |x| */
let mut y = f64::from_bits(u & 0x7fff_ffff_ffff_ffff);
if e < 0x3ff - 1 {
if e < 0x3ff - 32 {
/* handle underflow */
if e == 0 {
force_eval!(y as f32);
}
} else {
/* |x| < 0.5, up to 1.7ulp error */
y = 0.5 * log1p(2.0 * y + 2.0 * y * y / (1.0 - y));
}
} else {
/* avoid overflow */
y = 0.5 * log1p(2.0 * (y / (1.0 - y)));
}
if sign { -y } else { y }
}

View File

@ -0,0 +1,33 @@
use super::log1pf;
/* atanh(x) = log((1+x)/(1-x))/2 = log1p(2x/(1-x))/2 ~= x + x^3/3 + o(x^5) */
/// Inverse hyperbolic tangent (f32)
///
/// Calculates the inverse hyperbolic tangent of `x`.
/// Is defined as `log((1+x)/(1-x))/2 = log1p(2x/(1-x))/2`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn atanhf(mut x: f32) -> f32 {
let mut u = x.to_bits();
let sign = (u >> 31) != 0;
/* |x| */
u &= 0x7fffffff;
x = f32::from_bits(u);
if u < 0x3f800000 - (1 << 23) {
if u < 0x3f800000 - (32 << 23) {
/* handle underflow */
if u < (1 << 23) {
force_eval!((x * x) as f32);
}
} else {
/* |x| < 0.5, up to 1.7ulp error */
x = 0.5 * log1pf(2.0 * x + 2.0 * x * x / (1.0 - x));
}
} else {
/* avoid overflow */
x = 0.5 * log1pf(2.0 * (x / (1.0 - x)));
}
if sign { -x } else { x }
}

View File

@ -0,0 +1,113 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_cbrt.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*
* Optimized by Bruce D. Evans.
*/
/* cbrt(x)
* Return cube root of x
*/
use core::f64;
const B1: u32 = 715094163; /* B1 = (1023-1023/3-0.03306235651)*2**20 */
const B2: u32 = 696219795; /* B2 = (1023-1023/3-54/3-0.03306235651)*2**20 */
/* |1/cbrt(x) - p(x)| < 2**-23.5 (~[-7.93e-8, 7.929e-8]). */
const P0: f64 = 1.87595182427177009643; /* 0x3ffe03e6, 0x0f61e692 */
const P1: f64 = -1.88497979543377169875; /* 0xbffe28e0, 0x92f02420 */
const P2: f64 = 1.621429720105354466140; /* 0x3ff9f160, 0x4a49d6c2 */
const P3: f64 = -0.758397934778766047437; /* 0xbfe844cb, 0xbee751d9 */
const P4: f64 = 0.145996192886612446982; /* 0x3fc2b000, 0xd4e4edd7 */
// Cube root (f64)
///
/// Computes the cube root of the argument.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn cbrt(x: f64) -> f64 {
let x1p54 = f64::from_bits(0x4350000000000000); // 0x1p54 === 2 ^ 54
let mut ui: u64 = x.to_bits();
let mut r: f64;
let s: f64;
let mut t: f64;
let w: f64;
let mut hx: u32 = (ui >> 32) as u32 & 0x7fffffff;
if hx >= 0x7ff00000 {
/* cbrt(NaN,INF) is itself */
return x + x;
}
/*
* Rough cbrt to 5 bits:
* cbrt(2**e*(1+m) ~= 2**(e/3)*(1+(e%3+m)/3)
* where e is integral and >= 0, m is real and in [0, 1), and "/" and
* "%" are integer division and modulus with rounding towards minus
* infinity. The RHS is always >= the LHS and has a maximum relative
* error of about 1 in 16. Adding a bias of -0.03306235651 to the
* (e%3+m)/3 term reduces the error to about 1 in 32. With the IEEE
* floating point representation, for finite positive normal values,
* ordinary integer divison of the value in bits magically gives
* almost exactly the RHS of the above provided we first subtract the
* exponent bias (1023 for doubles) and later add it back. We do the
* subtraction virtually to keep e >= 0 so that ordinary integer
* division rounds towards minus infinity; this is also efficient.
*/
if hx < 0x00100000 {
/* zero or subnormal? */
ui = (x * x1p54).to_bits();
hx = (ui >> 32) as u32 & 0x7fffffff;
if hx == 0 {
return x; /* cbrt(0) is itself */
}
hx = hx / 3 + B2;
} else {
hx = hx / 3 + B1;
}
ui &= 1 << 63;
ui |= (hx as u64) << 32;
t = f64::from_bits(ui);
/*
* New cbrt to 23 bits:
* cbrt(x) = t*cbrt(x/t**3) ~= t*P(t**3/x)
* where P(r) is a polynomial of degree 4 that approximates 1/cbrt(r)
* to within 2**-23.5 when |r - 1| < 1/10. The rough approximation
* has produced t such than |t/cbrt(x) - 1| ~< 1/32, and cubing this
* gives us bounds for r = t**3/x.
*
* Try to optimize for parallel evaluation as in __tanf.c.
*/
r = (t * t) * (t / x);
t = t * ((P0 + r * (P1 + r * P2)) + ((r * r) * r) * (P3 + r * P4));
/*
* Round t away from zero to 23 bits (sloppily except for ensuring that
* the result is larger in magnitude than cbrt(x) but not much more than
* 2 23-bit ulps larger). With rounding towards zero, the error bound
* would be ~5/6 instead of ~4/6. With a maximum error of 2 23-bit ulps
* in the rounded t, the infinite-precision error in the Newton
* approximation barely affects third digit in the final error
* 0.667; the error in the rounded t can be up to about 3 23-bit ulps
* before the final error is larger than 0.667 ulps.
*/
ui = t.to_bits();
ui = (ui + 0x80000000) & 0xffffffffc0000000;
t = f64::from_bits(ui);
/* one step Newton iteration to 53 bits with error < 0.667 ulps */
s = t * t; /* t*t is exact */
r = x / s; /* error <= 0.5 ulps; |r| < |t| */
w = t + t; /* t+t is exact */
r = (r - t) / (w + r); /* r-t is exact; w+r ~= 3*t */
t = t + t * r; /* error <= 0.5 + 0.5/3 + epsilon */
t
}

View File

@ -0,0 +1,75 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_cbrtf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
* Debugged and optimized by Bruce D. Evans.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* cbrtf(x)
* Return cube root of x
*/
use core::f32;
const B1: u32 = 709958130; /* B1 = (127-127.0/3-0.03306235651)*2**23 */
const B2: u32 = 642849266; /* B2 = (127-127.0/3-24/3-0.03306235651)*2**23 */
/// Cube root (f32)
///
/// Computes the cube root of the argument.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn cbrtf(x: f32) -> f32 {
let x1p24 = f32::from_bits(0x4b800000); // 0x1p24f === 2 ^ 24
let mut r: f64;
let mut t: f64;
let mut ui: u32 = x.to_bits();
let mut hx: u32 = ui & 0x7fffffff;
if hx >= 0x7f800000 {
/* cbrt(NaN,INF) is itself */
return x + x;
}
/* rough cbrt to 5 bits */
if hx < 0x00800000 {
/* zero or subnormal? */
if hx == 0 {
return x; /* cbrt(+-0) is itself */
}
ui = (x * x1p24).to_bits();
hx = ui & 0x7fffffff;
hx = hx / 3 + B2;
} else {
hx = hx / 3 + B1;
}
ui &= 0x80000000;
ui |= hx;
/*
* First step Newton iteration (solving t*t-x/t == 0) to 16 bits. In
* double precision so that its terms can be arranged for efficiency
* without causing overflow or underflow.
*/
t = f32::from_bits(ui) as f64;
r = t * t * t;
t = t * (x as f64 + x as f64 + r) / (x as f64 + r + r);
/*
* Second step Newton iteration to 47 bits. In double precision for
* efficiency and accuracy.
*/
r = t * t * t;
t = t * (x as f64 + x as f64 + r) / (x as f64 + r + r);
/* rounding to 24 bits is perfect in round-to-nearest mode */
t as f32
}

View File

@ -0,0 +1,74 @@
#![allow(unreachable_code)]
use core::f64;
const TOINT: f64 = 1. / f64::EPSILON;
/// Ceil (f64)
///
/// Finds the nearest integer greater than or equal to `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ceil(x: f64) -> f64 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f64.ceil` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::ceilf64(x) }
}
}
#[cfg(all(target_arch = "x86", not(target_feature = "sse2")))]
{
//use an alternative implementation on x86, because the
//main implementation fails with the x87 FPU used by
//debian i386, probablly due to excess precision issues.
//basic implementation taken from https://github.com/rust-lang/libm/issues/219
use super::fabs;
if fabs(x).to_bits() < 4503599627370496.0_f64.to_bits() {
let truncated = x as i64 as f64;
if truncated < x {
return truncated + 1.0;
} else {
return truncated;
}
} else {
return x;
}
}
let u: u64 = x.to_bits();
let e: i64 = (u >> 52 & 0x7ff) as i64;
let y: f64;
if e >= 0x3ff + 52 || x == 0. {
return x;
}
// y = int(x) - x, where int(x) is an integer neighbor of x
y = if (u >> 63) != 0 { x - TOINT + TOINT - x } else { x + TOINT - TOINT - x };
// special case because of non-nearest rounding modes
if e < 0x3ff {
force_eval!(y);
return if (u >> 63) != 0 { -0. } else { 1. };
}
if y < 0. { x + y + 1. } else { x + y }
}
#[cfg(test)]
mod tests {
use super::*;
use core::f64::*;
#[test]
fn sanity_check() {
assert_eq!(ceil(1.1), 2.0);
assert_eq!(ceil(2.9), 3.0);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/ceil
#[test]
fn spec_tests() {
// Not Asserted: that the current rounding mode has no effect.
assert!(ceil(NAN).is_nan());
for f in [0.0, -0.0, INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(ceil(f), f);
}
}
}

View File

@ -0,0 +1,65 @@
use core::f32;
/// Ceil (f32)
///
/// Finds the nearest integer greater than or equal to `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ceilf(x: f32) -> f32 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f32.ceil` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::ceilf32(x) }
}
}
let mut ui = x.to_bits();
let e = (((ui >> 23) & 0xff).wrapping_sub(0x7f)) as i32;
if e >= 23 {
return x;
}
if e >= 0 {
let m = 0x007fffff >> e;
if (ui & m) == 0 {
return x;
}
force_eval!(x + f32::from_bits(0x7b800000));
if ui >> 31 == 0 {
ui += m;
}
ui &= !m;
} else {
force_eval!(x + f32::from_bits(0x7b800000));
if ui >> 31 != 0 {
return -0.0;
} else if ui << 1 != 0 {
return 1.0;
}
}
f32::from_bits(ui)
}
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
#[cfg(not(target_arch = "powerpc64"))]
#[cfg(test)]
mod tests {
use super::*;
use core::f32::*;
#[test]
fn sanity_check() {
assert_eq!(ceilf(1.1), 2.0);
assert_eq!(ceilf(2.9), 3.0);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/ceil
#[test]
fn spec_tests() {
// Not Asserted: that the current rounding mode has no effect.
assert!(ceilf(NAN).is_nan());
for f in [0.0, -0.0, INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(ceilf(f), f);
}
}
}

View File

@ -0,0 +1,12 @@
/// Sign of Y, magnitude of X (f64)
///
/// Constructs a number with the magnitude (absolute value) of its
/// first argument, `x`, and the sign of its second argument, `y`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn copysign(x: f64, y: f64) -> f64 {
let mut ux = x.to_bits();
let uy = y.to_bits();
ux &= (!0) >> 1;
ux |= uy & (1 << 63);
f64::from_bits(ux)
}

View File

@ -0,0 +1,12 @@
/// Sign of Y, magnitude of X (f32)
///
/// Constructs a number with the magnitude (absolute value) of its
/// first argument, `x`, and the sign of its second argument, `y`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn copysignf(x: f32, y: f32) -> f32 {
let mut ux = x.to_bits();
let uy = y.to_bits();
ux &= 0x7fffffff;
ux |= uy & 0x80000000;
f32::from_bits(ux)
}

View File

@ -0,0 +1,73 @@
// origin: FreeBSD /usr/src/lib/msun/src/s_cos.c */
//
// ====================================================
// Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
//
// Developed at SunPro, a Sun Microsystems, Inc. business.
// Permission to use, copy, modify, and distribute this
// software is freely granted, provided that this notice
// is preserved.
// ====================================================
use super::{k_cos, k_sin, rem_pio2};
// cos(x)
// Return cosine function of x.
//
// kernel function:
// k_sin ... sine function on [-pi/4,pi/4]
// k_cos ... cosine function on [-pi/4,pi/4]
// rem_pio2 ... argument reduction routine
//
// Method.
// Let S,C and T denote the sin, cos and tan respectively on
// [-PI/4, +PI/4]. Reduce the argument x to y1+y2 = x-k*pi/2
// in [-pi/4 , +pi/4], and let n = k mod 4.
// We have
//
// n sin(x) cos(x) tan(x)
// ----------------------------------------------------------
// 0 S C T
// 1 C -S -1/T
// 2 -S -C T
// 3 -C S -1/T
// ----------------------------------------------------------
//
// Special cases:
// Let trig be any of sin, cos, or tan.
// trig(+-INF) is NaN, with signals;
// trig(NaN) is that NaN;
//
// Accuracy:
// TRIG(x) returns trig(x) nearly rounded
//
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn cos(x: f64) -> f64 {
let ix = (f64::to_bits(x) >> 32) as u32 & 0x7fffffff;
/* |x| ~< pi/4 */
if ix <= 0x3fe921fb {
if ix < 0x3e46a09e {
/* if x < 2**-27 * sqrt(2) */
/* raise inexact if x != 0 */
if x as i32 == 0 {
return 1.0;
}
}
return k_cos(x, 0.0);
}
/* cos(Inf or NaN) is NaN */
if ix >= 0x7ff00000 {
return x - x;
}
/* argument reduction needed */
let (n, y0, y1) = rem_pio2(x);
match n & 3 {
0 => k_cos(y0, y1),
1 => -k_sin(y0, y1, 1),
2 => -k_cos(y0, y1),
_ => k_sin(y0, y1, 1),
}
}

View File

@ -0,0 +1,83 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_cosf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
* Optimized by Bruce D. Evans.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{k_cosf, k_sinf, rem_pio2f};
use core::f64::consts::FRAC_PI_2;
/* Small multiples of pi/2 rounded to double precision. */
const C1_PIO2: f64 = 1. * FRAC_PI_2; /* 0x3FF921FB, 0x54442D18 */
const C2_PIO2: f64 = 2. * FRAC_PI_2; /* 0x400921FB, 0x54442D18 */
const C3_PIO2: f64 = 3. * FRAC_PI_2; /* 0x4012D97C, 0x7F3321D2 */
const C4_PIO2: f64 = 4. * FRAC_PI_2; /* 0x401921FB, 0x54442D18 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn cosf(x: f32) -> f32 {
let x64 = x as f64;
let x1p120 = f32::from_bits(0x7b800000); // 0x1p120f === 2 ^ 120
let mut ix = x.to_bits();
let sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix <= 0x3f490fda {
/* |x| ~<= pi/4 */
if ix < 0x39800000 {
/* |x| < 2**-12 */
/* raise inexact if x != 0 */
force_eval!(x + x1p120);
return 1.;
}
return k_cosf(x64);
}
if ix <= 0x407b53d1 {
/* |x| ~<= 5*pi/4 */
if ix > 0x4016cbe3 {
/* |x| ~> 3*pi/4 */
return -k_cosf(if sign { x64 + C2_PIO2 } else { x64 - C2_PIO2 });
} else if sign {
return k_sinf(x64 + C1_PIO2);
} else {
return k_sinf(C1_PIO2 - x64);
}
}
if ix <= 0x40e231d5 {
/* |x| ~<= 9*pi/4 */
if ix > 0x40afeddf {
/* |x| ~> 7*pi/4 */
return k_cosf(if sign { x64 + C4_PIO2 } else { x64 - C4_PIO2 });
} else if sign {
return k_sinf(-x64 - C3_PIO2);
} else {
return k_sinf(x64 - C3_PIO2);
}
}
/* cos(Inf or NaN) is NaN */
if ix >= 0x7f800000 {
return x - x;
}
/* general argument reduction needed */
let (n, y) = rem_pio2f(x);
match n & 3 {
0 => k_cosf(y),
1 => k_sinf(-y),
2 => -k_cosf(y),
_ => k_sinf(y),
}
}

View File

@ -0,0 +1,38 @@
use super::exp;
use super::expm1;
use super::k_expo2;
/// Hyperbolic cosine (f64)
///
/// Computes the hyperbolic cosine of the argument x.
/// Is defined as `(exp(x) + exp(-x))/2`
/// Angles are specified in radians.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn cosh(mut x: f64) -> f64 {
/* |x| */
let mut ix = x.to_bits();
ix &= 0x7fffffffffffffff;
x = f64::from_bits(ix);
let w = ix >> 32;
/* |x| < log(2) */
if w < 0x3fe62e42 {
if w < 0x3ff00000 - (26 << 20) {
let x1p120 = f64::from_bits(0x4770000000000000);
force_eval!(x + x1p120);
return 1.;
}
let t = expm1(x); // exponential minus 1
return 1. + t * t / (2. * (1. + t));
}
/* |x| < log(DBL_MAX) */
if w < 0x40862e42 {
let t = exp(x);
/* note: if x>log(0x1p26) then the 1/t is not needed */
return 0.5 * (t + 1. / t);
}
/* |x| > log(DBL_MAX) or nan */
k_expo2(x)
}

View File

@ -0,0 +1,38 @@
use super::expf;
use super::expm1f;
use super::k_expo2f;
/// Hyperbolic cosine (f64)
///
/// Computes the hyperbolic cosine of the argument x.
/// Is defined as `(exp(x) + exp(-x))/2`
/// Angles are specified in radians.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn coshf(mut x: f32) -> f32 {
let x1p120 = f32::from_bits(0x7b800000); // 0x1p120f === 2 ^ 120
/* |x| */
let mut ix = x.to_bits();
ix &= 0x7fffffff;
x = f32::from_bits(ix);
let w = ix;
/* |x| < log(2) */
if w < 0x3f317217 {
if w < (0x3f800000 - (12 << 23)) {
force_eval!(x + x1p120);
return 1.;
}
let t = expm1f(x);
return 1. + t * t / (2. * (1. + t));
}
/* |x| < log(FLT_MAX) */
if w < 0x42b17217 {
let t = expf(x);
return 0.5 * (t + 1. / t);
}
/* |x| > log(FLT_MAX) or nan */
k_expo2f(x)
}

View File

@ -0,0 +1,310 @@
use super::{exp, fabs, get_high_word, with_set_low_word};
/* origin: FreeBSD /usr/src/lib/msun/src/s_erf.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* double erf(double x)
* double erfc(double x)
* x
* 2 |\
* erf(x) = --------- | exp(-t*t)dt
* sqrt(pi) \|
* 0
*
* erfc(x) = 1-erf(x)
* Note that
* erf(-x) = -erf(x)
* erfc(-x) = 2 - erfc(x)
*
* Method:
* 1. For |x| in [0, 0.84375]
* erf(x) = x + x*R(x^2)
* erfc(x) = 1 - erf(x) if x in [-.84375,0.25]
* = 0.5 + ((0.5-x)-x*R) if x in [0.25,0.84375]
* where R = P/Q where P is an odd poly of degree 8 and
* Q is an odd poly of degree 10.
* -57.90
* | R - (erf(x)-x)/x | <= 2
*
*
* Remark. The formula is derived by noting
* erf(x) = (2/sqrt(pi))*(x - x^3/3 + x^5/10 - x^7/42 + ....)
* and that
* 2/sqrt(pi) = 1.128379167095512573896158903121545171688
* is close to one. The interval is chosen because the fix
* point of erf(x) is near 0.6174 (i.e., erf(x)=x when x is
* near 0.6174), and by some experiment, 0.84375 is chosen to
* guarantee the error is less than one ulp for erf.
*
* 2. For |x| in [0.84375,1.25], let s = |x| - 1, and
* c = 0.84506291151 rounded to single (24 bits)
* erf(x) = sign(x) * (c + P1(s)/Q1(s))
* erfc(x) = (1-c) - P1(s)/Q1(s) if x > 0
* 1+(c+P1(s)/Q1(s)) if x < 0
* |P1/Q1 - (erf(|x|)-c)| <= 2**-59.06
* Remark: here we use the taylor series expansion at x=1.
* erf(1+s) = erf(1) + s*Poly(s)
* = 0.845.. + P1(s)/Q1(s)
* That is, we use rational approximation to approximate
* erf(1+s) - (c = (single)0.84506291151)
* Note that |P1/Q1|< 0.078 for x in [0.84375,1.25]
* where
* P1(s) = degree 6 poly in s
* Q1(s) = degree 6 poly in s
*
* 3. For x in [1.25,1/0.35(~2.857143)],
* erfc(x) = (1/x)*exp(-x*x-0.5625+R1/S1)
* erf(x) = 1 - erfc(x)
* where
* R1(z) = degree 7 poly in z, (z=1/x^2)
* S1(z) = degree 8 poly in z
*
* 4. For x in [1/0.35,28]
* erfc(x) = (1/x)*exp(-x*x-0.5625+R2/S2) if x > 0
* = 2.0 - (1/x)*exp(-x*x-0.5625+R2/S2) if -6<x<0
* = 2.0 - tiny (if x <= -6)
* erf(x) = sign(x)*(1.0 - erfc(x)) if x < 6, else
* erf(x) = sign(x)*(1.0 - tiny)
* where
* R2(z) = degree 6 poly in z, (z=1/x^2)
* S2(z) = degree 7 poly in z
*
* Note1:
* To compute exp(-x*x-0.5625+R/S), let s be a single
* precision number and s := x; then
* -x*x = -s*s + (s-x)*(s+x)
* exp(-x*x-0.5626+R/S) =
* exp(-s*s-0.5625)*exp((s-x)*(s+x)+R/S);
* Note2:
* Here 4 and 5 make use of the asymptotic series
* exp(-x*x)
* erfc(x) ~ ---------- * ( 1 + Poly(1/x^2) )
* x*sqrt(pi)
* We use rational approximation to approximate
* g(s)=f(1/x^2) = log(erfc(x)*x) - x*x + 0.5625
* Here is the error bound for R1/S1 and R2/S2
* |R1/S1 - f(x)| < 2**(-62.57)
* |R2/S2 - f(x)| < 2**(-61.52)
*
* 5. For inf > x >= 28
* erf(x) = sign(x) *(1 - tiny) (raise inexact)
* erfc(x) = tiny*tiny (raise underflow) if x > 0
* = 2 - tiny if x<0
*
* 7. Special case:
* erf(0) = 0, erf(inf) = 1, erf(-inf) = -1,
* erfc(0) = 1, erfc(inf) = 0, erfc(-inf) = 2,
* erfc/erf(NaN) is NaN
*/
const ERX: f64 = 8.45062911510467529297e-01; /* 0x3FEB0AC1, 0x60000000 */
/*
* Coefficients for approximation to erf on [0,0.84375]
*/
const EFX8: f64 = 1.02703333676410069053e+00; /* 0x3FF06EBA, 0x8214DB69 */
const PP0: f64 = 1.28379167095512558561e-01; /* 0x3FC06EBA, 0x8214DB68 */
const PP1: f64 = -3.25042107247001499370e-01; /* 0xBFD4CD7D, 0x691CB913 */
const PP2: f64 = -2.84817495755985104766e-02; /* 0xBF9D2A51, 0xDBD7194F */
const PP3: f64 = -5.77027029648944159157e-03; /* 0xBF77A291, 0x236668E4 */
const PP4: f64 = -2.37630166566501626084e-05; /* 0xBEF8EAD6, 0x120016AC */
const QQ1: f64 = 3.97917223959155352819e-01; /* 0x3FD97779, 0xCDDADC09 */
const QQ2: f64 = 6.50222499887672944485e-02; /* 0x3FB0A54C, 0x5536CEBA */
const QQ3: f64 = 5.08130628187576562776e-03; /* 0x3F74D022, 0xC4D36B0F */
const QQ4: f64 = 1.32494738004321644526e-04; /* 0x3F215DC9, 0x221C1A10 */
const QQ5: f64 = -3.96022827877536812320e-06; /* 0xBED09C43, 0x42A26120 */
/*
* Coefficients for approximation to erf in [0.84375,1.25]
*/
const PA0: f64 = -2.36211856075265944077e-03; /* 0xBF6359B8, 0xBEF77538 */
const PA1: f64 = 4.14856118683748331666e-01; /* 0x3FDA8D00, 0xAD92B34D */
const PA2: f64 = -3.72207876035701323847e-01; /* 0xBFD7D240, 0xFBB8C3F1 */
const PA3: f64 = 3.18346619901161753674e-01; /* 0x3FD45FCA, 0x805120E4 */
const PA4: f64 = -1.10894694282396677476e-01; /* 0xBFBC6398, 0x3D3E28EC */
const PA5: f64 = 3.54783043256182359371e-02; /* 0x3FA22A36, 0x599795EB */
const PA6: f64 = -2.16637559486879084300e-03; /* 0xBF61BF38, 0x0A96073F */
const QA1: f64 = 1.06420880400844228286e-01; /* 0x3FBB3E66, 0x18EEE323 */
const QA2: f64 = 5.40397917702171048937e-01; /* 0x3FE14AF0, 0x92EB6F33 */
const QA3: f64 = 7.18286544141962662868e-02; /* 0x3FB2635C, 0xD99FE9A7 */
const QA4: f64 = 1.26171219808761642112e-01; /* 0x3FC02660, 0xE763351F */
const QA5: f64 = 1.36370839120290507362e-02; /* 0x3F8BEDC2, 0x6B51DD1C */
const QA6: f64 = 1.19844998467991074170e-02; /* 0x3F888B54, 0x5735151D */
/*
* Coefficients for approximation to erfc in [1.25,1/0.35]
*/
const RA0: f64 = -9.86494403484714822705e-03; /* 0xBF843412, 0x600D6435 */
const RA1: f64 = -6.93858572707181764372e-01; /* 0xBFE63416, 0xE4BA7360 */
const RA2: f64 = -1.05586262253232909814e+01; /* 0xC0251E04, 0x41B0E726 */
const RA3: f64 = -6.23753324503260060396e+01; /* 0xC04F300A, 0xE4CBA38D */
const RA4: f64 = -1.62396669462573470355e+02; /* 0xC0644CB1, 0x84282266 */
const RA5: f64 = -1.84605092906711035994e+02; /* 0xC067135C, 0xEBCCABB2 */
const RA6: f64 = -8.12874355063065934246e+01; /* 0xC0545265, 0x57E4D2F2 */
const RA7: f64 = -9.81432934416914548592e+00; /* 0xC023A0EF, 0xC69AC25C */
const SA1: f64 = 1.96512716674392571292e+01; /* 0x4033A6B9, 0xBD707687 */
const SA2: f64 = 1.37657754143519042600e+02; /* 0x4061350C, 0x526AE721 */
const SA3: f64 = 4.34565877475229228821e+02; /* 0x407B290D, 0xD58A1A71 */
const SA4: f64 = 6.45387271733267880336e+02; /* 0x40842B19, 0x21EC2868 */
const SA5: f64 = 4.29008140027567833386e+02; /* 0x407AD021, 0x57700314 */
const SA6: f64 = 1.08635005541779435134e+02; /* 0x405B28A3, 0xEE48AE2C */
const SA7: f64 = 6.57024977031928170135e+00; /* 0x401A47EF, 0x8E484A93 */
const SA8: f64 = -6.04244152148580987438e-02; /* 0xBFAEEFF2, 0xEE749A62 */
/*
* Coefficients for approximation to erfc in [1/.35,28]
*/
const RB0: f64 = -9.86494292470009928597e-03; /* 0xBF843412, 0x39E86F4A */
const RB1: f64 = -7.99283237680523006574e-01; /* 0xBFE993BA, 0x70C285DE */
const RB2: f64 = -1.77579549177547519889e+01; /* 0xC031C209, 0x555F995A */
const RB3: f64 = -1.60636384855821916062e+02; /* 0xC064145D, 0x43C5ED98 */
const RB4: f64 = -6.37566443368389627722e+02; /* 0xC083EC88, 0x1375F228 */
const RB5: f64 = -1.02509513161107724954e+03; /* 0xC0900461, 0x6A2E5992 */
const RB6: f64 = -4.83519191608651397019e+02; /* 0xC07E384E, 0x9BDC383F */
const SB1: f64 = 3.03380607434824582924e+01; /* 0x403E568B, 0x261D5190 */
const SB2: f64 = 3.25792512996573918826e+02; /* 0x40745CAE, 0x221B9F0A */
const SB3: f64 = 1.53672958608443695994e+03; /* 0x409802EB, 0x189D5118 */
const SB4: f64 = 3.19985821950859553908e+03; /* 0x40A8FFB7, 0x688C246A */
const SB5: f64 = 2.55305040643316442583e+03; /* 0x40A3F219, 0xCEDF3BE6 */
const SB6: f64 = 4.74528541206955367215e+02; /* 0x407DA874, 0xE79FE763 */
const SB7: f64 = -2.24409524465858183362e+01; /* 0xC03670E2, 0x42712D62 */
fn erfc1(x: f64) -> f64 {
let s: f64;
let p: f64;
let q: f64;
s = fabs(x) - 1.0;
p = PA0 + s * (PA1 + s * (PA2 + s * (PA3 + s * (PA4 + s * (PA5 + s * PA6)))));
q = 1.0 + s * (QA1 + s * (QA2 + s * (QA3 + s * (QA4 + s * (QA5 + s * QA6)))));
1.0 - ERX - p / q
}
fn erfc2(ix: u32, mut x: f64) -> f64 {
let s: f64;
let r: f64;
let big_s: f64;
let z: f64;
if ix < 0x3ff40000 {
/* |x| < 1.25 */
return erfc1(x);
}
x = fabs(x);
s = 1.0 / (x * x);
if ix < 0x4006db6d {
/* |x| < 1/.35 ~ 2.85714 */
r = RA0 + s * (RA1 + s * (RA2 + s * (RA3 + s * (RA4 + s * (RA5 + s * (RA6 + s * RA7))))));
big_s = 1.0
+ s * (SA1
+ s * (SA2 + s * (SA3 + s * (SA4 + s * (SA5 + s * (SA6 + s * (SA7 + s * SA8)))))));
} else {
/* |x| > 1/.35 */
r = RB0 + s * (RB1 + s * (RB2 + s * (RB3 + s * (RB4 + s * (RB5 + s * RB6)))));
big_s =
1.0 + s * (SB1 + s * (SB2 + s * (SB3 + s * (SB4 + s * (SB5 + s * (SB6 + s * SB7))))));
}
z = with_set_low_word(x, 0);
exp(-z * z - 0.5625) * exp((z - x) * (z + x) + r / big_s) / x
}
/// Error function (f64)
///
/// Calculates an approximation to the “error function”, which estimates
/// the probability that an observation will fall within x standard
/// deviations of the mean (assuming a normal distribution).
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn erf(x: f64) -> f64 {
let r: f64;
let s: f64;
let z: f64;
let y: f64;
let mut ix: u32;
let sign: usize;
ix = get_high_word(x);
sign = (ix >> 31) as usize;
ix &= 0x7fffffff;
if ix >= 0x7ff00000 {
/* erf(nan)=nan, erf(+-inf)=+-1 */
return 1.0 - 2.0 * (sign as f64) + 1.0 / x;
}
if ix < 0x3feb0000 {
/* |x| < 0.84375 */
if ix < 0x3e300000 {
/* |x| < 2**-28 */
/* avoid underflow */
return 0.125 * (8.0 * x + EFX8 * x);
}
z = x * x;
r = PP0 + z * (PP1 + z * (PP2 + z * (PP3 + z * PP4)));
s = 1.0 + z * (QQ1 + z * (QQ2 + z * (QQ3 + z * (QQ4 + z * QQ5))));
y = r / s;
return x + x * y;
}
if ix < 0x40180000 {
/* 0.84375 <= |x| < 6 */
y = 1.0 - erfc2(ix, x);
} else {
let x1p_1022 = f64::from_bits(0x0010000000000000);
y = 1.0 - x1p_1022;
}
if sign != 0 { -y } else { y }
}
/// Complementary error function (f64)
///
/// Calculates the complementary probability.
/// Is `1 - erf(x)`. Is computed directly, so that you can use it to avoid
/// the loss of precision that would result from subtracting
/// large probabilities (on large `x`) from 1.
pub fn erfc(x: f64) -> f64 {
let r: f64;
let s: f64;
let z: f64;
let y: f64;
let mut ix: u32;
let sign: usize;
ix = get_high_word(x);
sign = (ix >> 31) as usize;
ix &= 0x7fffffff;
if ix >= 0x7ff00000 {
/* erfc(nan)=nan, erfc(+-inf)=0,2 */
return 2.0 * (sign as f64) + 1.0 / x;
}
if ix < 0x3feb0000 {
/* |x| < 0.84375 */
if ix < 0x3c700000 {
/* |x| < 2**-56 */
return 1.0 - x;
}
z = x * x;
r = PP0 + z * (PP1 + z * (PP2 + z * (PP3 + z * PP4)));
s = 1.0 + z * (QQ1 + z * (QQ2 + z * (QQ3 + z * (QQ4 + z * QQ5))));
y = r / s;
if sign != 0 || ix < 0x3fd00000 {
/* x < 1/4 */
return 1.0 - (x + x * y);
}
return 0.5 - (x - 0.5 + x * y);
}
if ix < 0x403c0000 {
/* 0.84375 <= |x| < 28 */
if sign != 0 {
return 2.0 - erfc2(ix, x);
} else {
return erfc2(ix, x);
}
}
let x1p_1022 = f64::from_bits(0x0010000000000000);
if sign != 0 { 2.0 - x1p_1022 } else { x1p_1022 * x1p_1022 }
}

View File

@ -0,0 +1,222 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_erff.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{expf, fabsf};
const ERX: f32 = 8.4506291151e-01; /* 0x3f58560b */
/*
* Coefficients for approximation to erf on [0,0.84375]
*/
const EFX8: f32 = 1.0270333290e+00; /* 0x3f8375d4 */
const PP0: f32 = 1.2837916613e-01; /* 0x3e0375d4 */
const PP1: f32 = -3.2504209876e-01; /* 0xbea66beb */
const PP2: f32 = -2.8481749818e-02; /* 0xbce9528f */
const PP3: f32 = -5.7702702470e-03; /* 0xbbbd1489 */
const PP4: f32 = -2.3763017452e-05; /* 0xb7c756b1 */
const QQ1: f32 = 3.9791721106e-01; /* 0x3ecbbbce */
const QQ2: f32 = 6.5022252500e-02; /* 0x3d852a63 */
const QQ3: f32 = 5.0813062117e-03; /* 0x3ba68116 */
const QQ4: f32 = 1.3249473704e-04; /* 0x390aee49 */
const QQ5: f32 = -3.9602282413e-06; /* 0xb684e21a */
/*
* Coefficients for approximation to erf in [0.84375,1.25]
*/
const PA0: f32 = -2.3621185683e-03; /* 0xbb1acdc6 */
const PA1: f32 = 4.1485610604e-01; /* 0x3ed46805 */
const PA2: f32 = -3.7220788002e-01; /* 0xbebe9208 */
const PA3: f32 = 3.1834661961e-01; /* 0x3ea2fe54 */
const PA4: f32 = -1.1089469492e-01; /* 0xbde31cc2 */
const PA5: f32 = 3.5478305072e-02; /* 0x3d1151b3 */
const PA6: f32 = -2.1663755178e-03; /* 0xbb0df9c0 */
const QA1: f32 = 1.0642088205e-01; /* 0x3dd9f331 */
const QA2: f32 = 5.4039794207e-01; /* 0x3f0a5785 */
const QA3: f32 = 7.1828655899e-02; /* 0x3d931ae7 */
const QA4: f32 = 1.2617121637e-01; /* 0x3e013307 */
const QA5: f32 = 1.3637083583e-02; /* 0x3c5f6e13 */
const QA6: f32 = 1.1984500103e-02; /* 0x3c445aa3 */
/*
* Coefficients for approximation to erfc in [1.25,1/0.35]
*/
const RA0: f32 = -9.8649440333e-03; /* 0xbc21a093 */
const RA1: f32 = -6.9385856390e-01; /* 0xbf31a0b7 */
const RA2: f32 = -1.0558626175e+01; /* 0xc128f022 */
const RA3: f32 = -6.2375331879e+01; /* 0xc2798057 */
const RA4: f32 = -1.6239666748e+02; /* 0xc322658c */
const RA5: f32 = -1.8460508728e+02; /* 0xc3389ae7 */
const RA6: f32 = -8.1287437439e+01; /* 0xc2a2932b */
const RA7: f32 = -9.8143291473e+00; /* 0xc11d077e */
const SA1: f32 = 1.9651271820e+01; /* 0x419d35ce */
const SA2: f32 = 1.3765776062e+02; /* 0x4309a863 */
const SA3: f32 = 4.3456588745e+02; /* 0x43d9486f */
const SA4: f32 = 6.4538726807e+02; /* 0x442158c9 */
const SA5: f32 = 4.2900814819e+02; /* 0x43d6810b */
const SA6: f32 = 1.0863500214e+02; /* 0x42d9451f */
const SA7: f32 = 6.5702495575e+00; /* 0x40d23f7c */
const SA8: f32 = -6.0424413532e-02; /* 0xbd777f97 */
/*
* Coefficients for approximation to erfc in [1/.35,28]
*/
const RB0: f32 = -9.8649431020e-03; /* 0xbc21a092 */
const RB1: f32 = -7.9928326607e-01; /* 0xbf4c9dd4 */
const RB2: f32 = -1.7757955551e+01; /* 0xc18e104b */
const RB3: f32 = -1.6063638306e+02; /* 0xc320a2ea */
const RB4: f32 = -6.3756646729e+02; /* 0xc41f6441 */
const RB5: f32 = -1.0250950928e+03; /* 0xc480230b */
const RB6: f32 = -4.8351919556e+02; /* 0xc3f1c275 */
const SB1: f32 = 3.0338060379e+01; /* 0x41f2b459 */
const SB2: f32 = 3.2579251099e+02; /* 0x43a2e571 */
const SB3: f32 = 1.5367296143e+03; /* 0x44c01759 */
const SB4: f32 = 3.1998581543e+03; /* 0x4547fdbb */
const SB5: f32 = 2.5530502930e+03; /* 0x451f90ce */
const SB6: f32 = 4.7452853394e+02; /* 0x43ed43a7 */
const SB7: f32 = -2.2440952301e+01; /* 0xc1b38712 */
fn erfc1(x: f32) -> f32 {
let s: f32;
let p: f32;
let q: f32;
s = fabsf(x) - 1.0;
p = PA0 + s * (PA1 + s * (PA2 + s * (PA3 + s * (PA4 + s * (PA5 + s * PA6)))));
q = 1.0 + s * (QA1 + s * (QA2 + s * (QA3 + s * (QA4 + s * (QA5 + s * QA6)))));
return 1.0 - ERX - p / q;
}
fn erfc2(mut ix: u32, mut x: f32) -> f32 {
let s: f32;
let r: f32;
let big_s: f32;
let z: f32;
if ix < 0x3fa00000 {
/* |x| < 1.25 */
return erfc1(x);
}
x = fabsf(x);
s = 1.0 / (x * x);
if ix < 0x4036db6d {
/* |x| < 1/0.35 */
r = RA0 + s * (RA1 + s * (RA2 + s * (RA3 + s * (RA4 + s * (RA5 + s * (RA6 + s * RA7))))));
big_s = 1.0
+ s * (SA1
+ s * (SA2 + s * (SA3 + s * (SA4 + s * (SA5 + s * (SA6 + s * (SA7 + s * SA8)))))));
} else {
/* |x| >= 1/0.35 */
r = RB0 + s * (RB1 + s * (RB2 + s * (RB3 + s * (RB4 + s * (RB5 + s * RB6)))));
big_s =
1.0 + s * (SB1 + s * (SB2 + s * (SB3 + s * (SB4 + s * (SB5 + s * (SB6 + s * SB7))))));
}
ix = x.to_bits();
z = f32::from_bits(ix & 0xffffe000);
expf(-z * z - 0.5625) * expf((z - x) * (z + x) + r / big_s) / x
}
/// Error function (f32)
///
/// Calculates an approximation to the “error function”, which estimates
/// the probability that an observation will fall within x standard
/// deviations of the mean (assuming a normal distribution).
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn erff(x: f32) -> f32 {
let r: f32;
let s: f32;
let z: f32;
let y: f32;
let mut ix: u32;
let sign: usize;
ix = x.to_bits();
sign = (ix >> 31) as usize;
ix &= 0x7fffffff;
if ix >= 0x7f800000 {
/* erf(nan)=nan, erf(+-inf)=+-1 */
return 1.0 - 2.0 * (sign as f32) + 1.0 / x;
}
if ix < 0x3f580000 {
/* |x| < 0.84375 */
if ix < 0x31800000 {
/* |x| < 2**-28 */
/*avoid underflow */
return 0.125 * (8.0 * x + EFX8 * x);
}
z = x * x;
r = PP0 + z * (PP1 + z * (PP2 + z * (PP3 + z * PP4)));
s = 1.0 + z * (QQ1 + z * (QQ2 + z * (QQ3 + z * (QQ4 + z * QQ5))));
y = r / s;
return x + x * y;
}
if ix < 0x40c00000 {
/* |x| < 6 */
y = 1.0 - erfc2(ix, x);
} else {
let x1p_120 = f32::from_bits(0x03800000);
y = 1.0 - x1p_120;
}
if sign != 0 { -y } else { y }
}
/// Complementary error function (f32)
///
/// Calculates the complementary probability.
/// Is `1 - erf(x)`. Is computed directly, so that you can use it to avoid
/// the loss of precision that would result from subtracting
/// large probabilities (on large `x`) from 1.
pub fn erfcf(x: f32) -> f32 {
let r: f32;
let s: f32;
let z: f32;
let y: f32;
let mut ix: u32;
let sign: usize;
ix = x.to_bits();
sign = (ix >> 31) as usize;
ix &= 0x7fffffff;
if ix >= 0x7f800000 {
/* erfc(nan)=nan, erfc(+-inf)=0,2 */
return 2.0 * (sign as f32) + 1.0 / x;
}
if ix < 0x3f580000 {
/* |x| < 0.84375 */
if ix < 0x23800000 {
/* |x| < 2**-56 */
return 1.0 - x;
}
z = x * x;
r = PP0 + z * (PP1 + z * (PP2 + z * (PP3 + z * PP4)));
s = 1.0 + z * (QQ1 + z * (QQ2 + z * (QQ3 + z * (QQ4 + z * QQ5))));
y = r / s;
if sign != 0 || ix < 0x3e800000 {
/* x < 1/4 */
return 1.0 - (x + x * y);
}
return 0.5 - (x - 0.5 + x * y);
}
if ix < 0x41e00000 {
/* |x| < 28 */
if sign != 0 {
return 2.0 - erfc2(ix, x);
} else {
return erfc2(ix, x);
}
}
let x1p_120 = f32::from_bits(0x03800000);
if sign != 0 { 2.0 - x1p_120 } else { x1p_120 * x1p_120 }
}

View File

@ -0,0 +1,150 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_exp.c */
/*
* ====================================================
* Copyright (C) 2004 by Sun Microsystems, Inc. All rights reserved.
*
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* exp(x)
* Returns the exponential of x.
*
* Method
* 1. Argument reduction:
* Reduce x to an r so that |r| <= 0.5*ln2 ~ 0.34658.
* Given x, find r and integer k such that
*
* x = k*ln2 + r, |r| <= 0.5*ln2.
*
* Here r will be represented as r = hi-lo for better
* accuracy.
*
* 2. Approximation of exp(r) by a special rational function on
* the interval [0,0.34658]:
* Write
* R(r**2) = r*(exp(r)+1)/(exp(r)-1) = 2 + r*r/6 - r**4/360 + ...
* We use a special Remez algorithm on [0,0.34658] to generate
* a polynomial of degree 5 to approximate R. The maximum error
* of this polynomial approximation is bounded by 2**-59. In
* other words,
* R(z) ~ 2.0 + P1*z + P2*z**2 + P3*z**3 + P4*z**4 + P5*z**5
* (where z=r*r, and the values of P1 to P5 are listed below)
* and
* | 5 | -59
* | 2.0+P1*z+...+P5*z - R(z) | <= 2
* | |
* The computation of exp(r) thus becomes
* 2*r
* exp(r) = 1 + ----------
* R(r) - r
* r*c(r)
* = 1 + r + ----------- (for better accuracy)
* 2 - c(r)
* where
* 2 4 10
* c(r) = r - (P1*r + P2*r + ... + P5*r ).
*
* 3. Scale back to obtain exp(x):
* From step 1, we have
* exp(x) = 2^k * exp(r)
*
* Special cases:
* exp(INF) is INF, exp(NaN) is NaN;
* exp(-INF) is 0, and
* for finite argument, only exp(0)=1 is exact.
*
* Accuracy:
* according to an error analysis, the error is always less than
* 1 ulp (unit in the last place).
*
* Misc. info.
* For IEEE double
* if x > 709.782712893383973096 then exp(x) overflows
* if x < -745.133219101941108420 then exp(x) underflows
*/
use super::scalbn;
const HALF: [f64; 2] = [0.5, -0.5];
const LN2HI: f64 = 6.93147180369123816490e-01; /* 0x3fe62e42, 0xfee00000 */
const LN2LO: f64 = 1.90821492927058770002e-10; /* 0x3dea39ef, 0x35793c76 */
const INVLN2: f64 = 1.44269504088896338700e+00; /* 0x3ff71547, 0x652b82fe */
const P1: f64 = 1.66666666666666019037e-01; /* 0x3FC55555, 0x5555553E */
const P2: f64 = -2.77777777770155933842e-03; /* 0xBF66C16C, 0x16BEBD93 */
const P3: f64 = 6.61375632143793436117e-05; /* 0x3F11566A, 0xAF25DE2C */
const P4: f64 = -1.65339022054652515390e-06; /* 0xBEBBBD41, 0xC5D26BF1 */
const P5: f64 = 4.13813679705723846039e-08; /* 0x3E663769, 0x72BEA4D0 */
/// Exponential, base *e* (f64)
///
/// Calculate the exponential of `x`, that is, *e* raised to the power `x`
/// (where *e* is the base of the natural system of logarithms, approximately 2.71828).
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn exp(mut x: f64) -> f64 {
let x1p1023 = f64::from_bits(0x7fe0000000000000); // 0x1p1023 === 2 ^ 1023
let x1p_149 = f64::from_bits(0x36a0000000000000); // 0x1p-149 === 2 ^ -149
let hi: f64;
let lo: f64;
let c: f64;
let xx: f64;
let y: f64;
let k: i32;
let sign: i32;
let mut hx: u32;
hx = (x.to_bits() >> 32) as u32;
sign = (hx >> 31) as i32;
hx &= 0x7fffffff; /* high word of |x| */
/* special cases */
if hx >= 0x4086232b {
/* if |x| >= 708.39... */
if x.is_nan() {
return x;
}
if x > 709.782712893383973096 {
/* overflow if x!=inf */
x *= x1p1023;
return x;
}
if x < -708.39641853226410622 {
/* underflow if x!=-inf */
force_eval!((-x1p_149 / x) as f32);
if x < -745.13321910194110842 {
return 0.;
}
}
}
/* argument reduction */
if hx > 0x3fd62e42 {
/* if |x| > 0.5 ln2 */
if hx >= 0x3ff0a2b2 {
/* if |x| >= 1.5 ln2 */
k = (INVLN2 * x + i!(HALF, sign as usize)) as i32;
} else {
k = 1 - sign - sign;
}
hi = x - k as f64 * LN2HI; /* k*ln2hi is exact here */
lo = k as f64 * LN2LO;
x = hi - lo;
} else if hx > 0x3e300000 {
/* if |x| > 2**-28 */
k = 0;
hi = x;
lo = 0.;
} else {
/* inexact if x!=0 */
force_eval!(x1p1023 + x);
return 1. + x;
}
/* x is now in primary range */
xx = x * x;
c = x - xx * (P1 + xx * (P2 + xx * (P3 + xx * (P4 + xx * P5))));
y = 1. + (x * c / (2. - c) - lo + hi);
if k == 0 { y } else { scalbn(y, k) }
}

View File

@ -0,0 +1,22 @@
use super::{exp2, modf, pow};
const LN10: f64 = 3.32192809488736234787031942948939;
const P10: &[f64] = &[
1e-15, 1e-14, 1e-13, 1e-12, 1e-11, 1e-10, 1e-9, 1e-8, 1e-7, 1e-6, 1e-5, 1e-4, 1e-3, 1e-2, 1e-1,
1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9, 1e10, 1e11, 1e12, 1e13, 1e14, 1e15,
];
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn exp10(x: f64) -> f64 {
let (mut y, n) = modf(x);
let u: u64 = n.to_bits();
/* fabs(n) < 16 without raising invalid on nan */
if (u >> 52 & 0x7ff) < 0x3ff + 4 {
if y == 0.0 {
return i!(P10, ((n as isize) + 15) as usize);
}
y = exp2(LN10 * y);
return y * i!(P10, ((n as isize) + 15) as usize);
}
return pow(10.0, x);
}

View File

@ -0,0 +1,21 @@
use super::{exp2, exp2f, modff};
const LN10_F32: f32 = 3.32192809488736234787031942948939;
const LN10_F64: f64 = 3.32192809488736234787031942948939;
const P10: &[f32] =
&[1e-7, 1e-6, 1e-5, 1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7];
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn exp10f(x: f32) -> f32 {
let (mut y, n) = modff(x);
let u = n.to_bits();
/* fabsf(n) < 8 without raising invalid on nan */
if (u >> 23 & 0xff) < 0x7f + 3 {
if y == 0.0 {
return i!(P10, ((n as isize) + 7) as usize);
}
y = exp2f(LN10_F32 * y);
return y * i!(P10, ((n as isize) + 7) as usize);
}
return exp2(LN10_F64 * (x as f64)) as f32;
}

View File

@ -0,0 +1,394 @@
// origin: FreeBSD /usr/src/lib/msun/src/s_exp2.c */
//-
// Copyright (c) 2005 David Schultz <das@FreeBSD.ORG>
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
// 1. Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// 2. Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
// ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
// OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
// HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
// OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
// SUCH DAMAGE.
use super::scalbn;
const TBLSIZE: usize = 256;
#[cfg_attr(rustfmt, rustfmt_skip)]
static TBL: [u64; TBLSIZE * 2] = [
// exp2(z + eps) eps
0x3fe6a09e667f3d5d, 0x3d39880000000000,
0x3fe6b052fa751744, 0x3cd8000000000000,
0x3fe6c012750bd9fe, 0xbd28780000000000,
0x3fe6cfdcddd476bf, 0x3d1ec00000000000,
0x3fe6dfb23c651a29, 0xbcd8000000000000,
0x3fe6ef9298593ae3, 0xbcbc000000000000,
0x3fe6ff7df9519386, 0xbd2fd80000000000,
0x3fe70f7466f42da3, 0xbd2c880000000000,
0x3fe71f75e8ec5fc3, 0x3d13c00000000000,
0x3fe72f8286eacf05, 0xbd38300000000000,
0x3fe73f9a48a58152, 0xbd00c00000000000,
0x3fe74fbd35d7ccfc, 0x3d2f880000000000,
0x3fe75feb564267f1, 0x3d03e00000000000,
0x3fe77024b1ab6d48, 0xbd27d00000000000,
0x3fe780694fde5d38, 0xbcdd000000000000,
0x3fe790b938ac1d00, 0x3ce3000000000000,
0x3fe7a11473eb0178, 0xbced000000000000,
0x3fe7b17b0976d060, 0x3d20400000000000,
0x3fe7c1ed0130c133, 0x3ca0000000000000,
0x3fe7d26a62ff8636, 0xbd26900000000000,
0x3fe7e2f336cf4e3b, 0xbd02e00000000000,
0x3fe7f3878491c3e8, 0xbd24580000000000,
0x3fe80427543e1b4e, 0x3d33000000000000,
0x3fe814d2add1071a, 0x3d0f000000000000,
0x3fe82589994ccd7e, 0xbd21c00000000000,
0x3fe8364c1eb942d0, 0x3d29d00000000000,
0x3fe8471a4623cab5, 0x3d47100000000000,
0x3fe857f4179f5bbc, 0x3d22600000000000,
0x3fe868d99b4491af, 0xbd32c40000000000,
0x3fe879cad931a395, 0xbd23000000000000,
0x3fe88ac7d98a65b8, 0xbd2a800000000000,
0x3fe89bd0a4785800, 0xbced000000000000,
0x3fe8ace5422aa223, 0x3d33280000000000,
0x3fe8be05bad619fa, 0x3d42b40000000000,
0x3fe8cf3216b54383, 0xbd2ed00000000000,
0x3fe8e06a5e08664c, 0xbd20500000000000,
0x3fe8f1ae99157807, 0x3d28280000000000,
0x3fe902fed0282c0e, 0xbd1cb00000000000,
0x3fe9145b0b91ff96, 0xbd05e00000000000,
0x3fe925c353aa2ff9, 0x3cf5400000000000,
0x3fe93737b0cdc64a, 0x3d17200000000000,
0x3fe948b82b5f98ae, 0xbd09000000000000,
0x3fe95a44cbc852cb, 0x3d25680000000000,
0x3fe96bdd9a766f21, 0xbd36d00000000000,
0x3fe97d829fde4e2a, 0xbd01000000000000,
0x3fe98f33e47a23a3, 0x3d2d000000000000,
0x3fe9a0f170ca0604, 0xbd38a40000000000,
0x3fe9b2bb4d53ff89, 0x3d355c0000000000,
0x3fe9c49182a3f15b, 0x3d26b80000000000,
0x3fe9d674194bb8c5, 0xbcec000000000000,
0x3fe9e86319e3238e, 0x3d17d00000000000,
0x3fe9fa5e8d07f302, 0x3d16400000000000,
0x3fea0c667b5de54d, 0xbcf5000000000000,
0x3fea1e7aed8eb8f6, 0x3d09e00000000000,
0x3fea309bec4a2e27, 0x3d2ad80000000000,
0x3fea42c980460a5d, 0xbd1af00000000000,
0x3fea5503b23e259b, 0x3d0b600000000000,
0x3fea674a8af46213, 0x3d38880000000000,
0x3fea799e1330b3a7, 0x3d11200000000000,
0x3fea8bfe53c12e8d, 0x3d06c00000000000,
0x3fea9e6b5579fcd2, 0xbd29b80000000000,
0x3feab0e521356fb8, 0x3d2b700000000000,
0x3feac36bbfd3f381, 0x3cd9000000000000,
0x3fead5ff3a3c2780, 0x3ce4000000000000,
0x3feae89f995ad2a3, 0xbd2c900000000000,
0x3feafb4ce622f367, 0x3d16500000000000,
0x3feb0e07298db790, 0x3d2fd40000000000,
0x3feb20ce6c9a89a9, 0x3d12700000000000,
0x3feb33a2b84f1a4b, 0x3d4d470000000000,
0x3feb468415b747e7, 0xbd38380000000000,
0x3feb59728de5593a, 0x3c98000000000000,
0x3feb6c6e29f1c56a, 0x3d0ad00000000000,
0x3feb7f76f2fb5e50, 0x3cde800000000000,
0x3feb928cf22749b2, 0xbd04c00000000000,
0x3feba5b030a10603, 0xbd0d700000000000,
0x3febb8e0b79a6f66, 0x3d0d900000000000,
0x3febcc1e904bc1ff, 0x3d02a00000000000,
0x3febdf69c3f3a16f, 0xbd1f780000000000,
0x3febf2c25bd71db8, 0xbd10a00000000000,
0x3fec06286141b2e9, 0xbd11400000000000,
0x3fec199bdd8552e0, 0x3d0be00000000000,
0x3fec2d1cd9fa64ee, 0xbd09400000000000,
0x3fec40ab5fffd02f, 0xbd0ed00000000000,
0x3fec544778fafd15, 0x3d39660000000000,
0x3fec67f12e57d0cb, 0xbd1a100000000000,
0x3fec7ba88988c1b6, 0xbd58458000000000,
0x3fec8f6d9406e733, 0xbd1a480000000000,
0x3feca3405751c4df, 0x3ccb000000000000,
0x3fecb720dcef9094, 0x3d01400000000000,
0x3feccb0f2e6d1689, 0x3cf0200000000000,
0x3fecdf0b555dc412, 0x3cf3600000000000,
0x3fecf3155b5bab3b, 0xbd06900000000000,
0x3fed072d4a0789bc, 0x3d09a00000000000,
0x3fed1b532b08c8fa, 0xbd15e00000000000,
0x3fed2f87080d8a85, 0x3d1d280000000000,
0x3fed43c8eacaa203, 0x3d01a00000000000,
0x3fed5818dcfba491, 0x3cdf000000000000,
0x3fed6c76e862e6a1, 0xbd03a00000000000,
0x3fed80e316c9834e, 0xbd0cd80000000000,
0x3fed955d71ff6090, 0x3cf4c00000000000,
0x3feda9e603db32ae, 0x3cff900000000000,
0x3fedbe7cd63a8325, 0x3ce9800000000000,
0x3fedd321f301b445, 0xbcf5200000000000,
0x3fede7d5641c05bf, 0xbd1d700000000000,
0x3fedfc97337b9aec, 0xbd16140000000000,
0x3fee11676b197d5e, 0x3d0b480000000000,
0x3fee264614f5a3e7, 0x3d40ce0000000000,
0x3fee3b333b16ee5c, 0x3d0c680000000000,
0x3fee502ee78b3fb4, 0xbd09300000000000,
0x3fee653924676d68, 0xbce5000000000000,
0x3fee7a51fbc74c44, 0xbd07f80000000000,
0x3fee8f7977cdb726, 0xbcf3700000000000,
0x3feea4afa2a490e8, 0x3ce5d00000000000,
0x3feeb9f4867ccae4, 0x3d161a0000000000,
0x3feecf482d8e680d, 0x3cf5500000000000,
0x3feee4aaa2188514, 0x3cc6400000000000,
0x3feefa1bee615a13, 0xbcee800000000000,
0x3fef0f9c1cb64106, 0xbcfa880000000000,
0x3fef252b376bb963, 0xbd2c900000000000,
0x3fef3ac948dd7275, 0x3caa000000000000,
0x3fef50765b6e4524, 0xbcf4f00000000000,
0x3fef6632798844fd, 0x3cca800000000000,
0x3fef7bfdad9cbe38, 0x3cfabc0000000000,
0x3fef91d802243c82, 0xbcd4600000000000,
0x3fefa7c1819e908e, 0xbd0b0c0000000000,
0x3fefbdba3692d511, 0xbcc0e00000000000,
0x3fefd3c22b8f7194, 0xbd10de8000000000,
0x3fefe9d96b2a23ee, 0x3cee430000000000,
0x3ff0000000000000, 0x0,
0x3ff00b1afa5abcbe, 0xbcb3400000000000,
0x3ff0163da9fb3303, 0xbd12170000000000,
0x3ff02168143b0282, 0x3cba400000000000,
0x3ff02c9a3e77806c, 0x3cef980000000000,
0x3ff037d42e11bbca, 0xbcc7400000000000,
0x3ff04315e86e7f89, 0x3cd8300000000000,
0x3ff04e5f72f65467, 0xbd1a3f0000000000,
0x3ff059b0d315855a, 0xbd02840000000000,
0x3ff0650a0e3c1f95, 0x3cf1600000000000,
0x3ff0706b29ddf71a, 0x3d15240000000000,
0x3ff07bd42b72a82d, 0xbce9a00000000000,
0x3ff0874518759bd0, 0x3ce6400000000000,
0x3ff092bdf66607c8, 0xbd00780000000000,
0x3ff09e3ecac6f383, 0xbc98000000000000,
0x3ff0a9c79b1f3930, 0x3cffa00000000000,
0x3ff0b5586cf988fc, 0xbcfac80000000000,
0x3ff0c0f145e46c8a, 0x3cd9c00000000000,
0x3ff0cc922b724816, 0x3d05200000000000,
0x3ff0d83b23395dd8, 0xbcfad00000000000,
0x3ff0e3ec32d3d1f3, 0x3d1bac0000000000,
0x3ff0efa55fdfa9a6, 0xbd04e80000000000,
0x3ff0fb66affed2f0, 0xbd0d300000000000,
0x3ff1073028d7234b, 0x3cf1500000000000,
0x3ff11301d0125b5b, 0x3cec000000000000,
0x3ff11edbab5e2af9, 0x3d16bc0000000000,
0x3ff12abdc06c31d5, 0x3ce8400000000000,
0x3ff136a814f2047d, 0xbd0ed00000000000,
0x3ff1429aaea92de9, 0x3ce8e00000000000,
0x3ff14e95934f3138, 0x3ceb400000000000,
0x3ff15a98c8a58e71, 0x3d05300000000000,
0x3ff166a45471c3df, 0x3d03380000000000,
0x3ff172b83c7d5211, 0x3d28d40000000000,
0x3ff17ed48695bb9f, 0xbd05d00000000000,
0x3ff18af9388c8d93, 0xbd1c880000000000,
0x3ff1972658375d66, 0x3d11f00000000000,
0x3ff1a35beb6fcba7, 0x3d10480000000000,
0x3ff1af99f81387e3, 0xbd47390000000000,
0x3ff1bbe084045d54, 0x3d24e40000000000,
0x3ff1c82f95281c43, 0xbd0a200000000000,
0x3ff1d4873168b9b2, 0x3ce3800000000000,
0x3ff1e0e75eb44031, 0x3ceac00000000000,
0x3ff1ed5022fcd938, 0x3d01900000000000,
0x3ff1f9c18438cdf7, 0xbd1b780000000000,
0x3ff2063b88628d8f, 0x3d2d940000000000,
0x3ff212be3578a81e, 0x3cd8000000000000,
0x3ff21f49917ddd41, 0x3d2b340000000000,
0x3ff22bdda2791323, 0x3d19f80000000000,
0x3ff2387a6e7561e7, 0xbd19c80000000000,
0x3ff2451ffb821427, 0x3d02300000000000,
0x3ff251ce4fb2a602, 0xbd13480000000000,
0x3ff25e85711eceb0, 0x3d12700000000000,
0x3ff26b4565e27d16, 0x3d11d00000000000,
0x3ff2780e341de00f, 0x3d31ee0000000000,
0x3ff284dfe1f5633e, 0xbd14c00000000000,
0x3ff291ba7591bb30, 0xbd13d80000000000,
0x3ff29e9df51fdf09, 0x3d08b00000000000,
0x3ff2ab8a66d10e9b, 0xbd227c0000000000,
0x3ff2b87fd0dada3a, 0x3d2a340000000000,
0x3ff2c57e39771af9, 0xbd10800000000000,
0x3ff2d285a6e402d9, 0xbd0ed00000000000,
0x3ff2df961f641579, 0xbcf4200000000000,
0x3ff2ecafa93e2ecf, 0xbd24980000000000,
0x3ff2f9d24abd8822, 0xbd16300000000000,
0x3ff306fe0a31b625, 0xbd32360000000000,
0x3ff31432edeea50b, 0xbd70df8000000000,
0x3ff32170fc4cd7b8, 0xbd22480000000000,
0x3ff32eb83ba8e9a2, 0xbd25980000000000,
0x3ff33c08b2641766, 0x3d1ed00000000000,
0x3ff3496266e3fa27, 0xbcdc000000000000,
0x3ff356c55f929f0f, 0xbd30d80000000000,
0x3ff36431a2de88b9, 0x3d22c80000000000,
0x3ff371a7373aaa39, 0x3d20600000000000,
0x3ff37f26231e74fe, 0xbd16600000000000,
0x3ff38cae6d05d838, 0xbd0ae00000000000,
0x3ff39a401b713ec3, 0xbd44720000000000,
0x3ff3a7db34e5a020, 0x3d08200000000000,
0x3ff3b57fbfec6e95, 0x3d3e800000000000,
0x3ff3c32dc313a8f2, 0x3cef800000000000,
0x3ff3d0e544ede122, 0xbd17a00000000000,
0x3ff3dea64c1234bb, 0x3d26300000000000,
0x3ff3ec70df1c4ecc, 0xbd48a60000000000,
0x3ff3fa4504ac7e8c, 0xbd3cdc0000000000,
0x3ff40822c367a0bb, 0x3d25b80000000000,
0x3ff4160a21f72e95, 0x3d1ec00000000000,
0x3ff423fb27094646, 0xbd13600000000000,
0x3ff431f5d950a920, 0x3d23980000000000,
0x3ff43ffa3f84b9eb, 0x3cfa000000000000,
0x3ff44e0860618919, 0xbcf6c00000000000,
0x3ff45c2042a7d201, 0xbd0bc00000000000,
0x3ff46a41ed1d0016, 0xbd12800000000000,
0x3ff4786d668b3326, 0x3d30e00000000000,
0x3ff486a2b5c13c00, 0xbd2d400000000000,
0x3ff494e1e192af04, 0x3d0c200000000000,
0x3ff4a32af0d7d372, 0xbd1e500000000000,
0x3ff4b17dea6db801, 0x3d07800000000000,
0x3ff4bfdad53629e1, 0xbd13800000000000,
0x3ff4ce41b817c132, 0x3d00800000000000,
0x3ff4dcb299fddddb, 0x3d2c700000000000,
0x3ff4eb2d81d8ab96, 0xbd1ce00000000000,
0x3ff4f9b2769d2d02, 0x3d19200000000000,
0x3ff508417f4531c1, 0xbd08c00000000000,
0x3ff516daa2cf662a, 0xbcfa000000000000,
0x3ff5257de83f51ea, 0x3d4a080000000000,
0x3ff5342b569d4eda, 0xbd26d80000000000,
0x3ff542e2f4f6ac1a, 0xbd32440000000000,
0x3ff551a4ca5d94db, 0x3d483c0000000000,
0x3ff56070dde9116b, 0x3d24b00000000000,
0x3ff56f4736b529de, 0x3d415a0000000000,
0x3ff57e27dbe2c40e, 0xbd29e00000000000,
0x3ff58d12d497c76f, 0xbd23080000000000,
0x3ff59c0827ff0b4c, 0x3d4dec0000000000,
0x3ff5ab07dd485427, 0xbcc4000000000000,
0x3ff5ba11fba87af4, 0x3d30080000000000,
0x3ff5c9268a59460b, 0xbd26c80000000000,
0x3ff5d84590998e3f, 0x3d469a0000000000,
0x3ff5e76f15ad20e1, 0xbd1b400000000000,
0x3ff5f6a320dcebca, 0x3d17700000000000,
0x3ff605e1b976dcb8, 0x3d26f80000000000,
0x3ff6152ae6cdf715, 0x3d01000000000000,
0x3ff6247eb03a5531, 0xbd15d00000000000,
0x3ff633dd1d1929b5, 0xbd12d00000000000,
0x3ff6434634ccc313, 0xbcea800000000000,
0x3ff652b9febc8efa, 0xbd28600000000000,
0x3ff6623882553397, 0x3d71fe0000000000,
0x3ff671c1c708328e, 0xbd37200000000000,
0x3ff68155d44ca97e, 0x3ce6800000000000,
0x3ff690f4b19e9471, 0xbd29780000000000,
];
// exp2(x): compute the base 2 exponential of x
//
// Accuracy: Peak error < 0.503 ulp for normalized results.
//
// Method: (accurate tables)
//
// Reduce x:
// x = k + y, for integer k and |y| <= 1/2.
// Thus we have exp2(x) = 2**k * exp2(y).
//
// Reduce y:
// y = i/TBLSIZE + z - eps[i] for integer i near y * TBLSIZE.
// Thus we have exp2(y) = exp2(i/TBLSIZE) * exp2(z - eps[i]),
// with |z - eps[i]| <= 2**-9 + 2**-39 for the table used.
//
// We compute exp2(i/TBLSIZE) via table lookup and exp2(z - eps[i]) via
// a degree-5 minimax polynomial with maximum error under 1.3 * 2**-61.
// The values in exp2t[] and eps[] are chosen such that
// exp2t[i] = exp2(i/TBLSIZE + eps[i]), and eps[i] is a small offset such
// that exp2t[i] is accurate to 2**-64.
//
// Note that the range of i is +-TBLSIZE/2, so we actually index the tables
// by i0 = i + TBLSIZE/2. For cache efficiency, exp2t[] and eps[] are
// virtual tables, interleaved in the real table tbl[].
//
// This method is due to Gal, with many details due to Gal and Bachelis:
//
// Gal, S. and Bachelis, B. An Accurate Elementary Mathematical Library
// for the IEEE Floating Point Standard. TOMS 17(1), 26-46 (1991).
/// Exponential, base 2 (f64)
///
/// Calculate `2^x`, that is, 2 raised to the power `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn exp2(mut x: f64) -> f64 {
let redux = f64::from_bits(0x4338000000000000) / TBLSIZE as f64;
let p1 = f64::from_bits(0x3fe62e42fefa39ef);
let p2 = f64::from_bits(0x3fcebfbdff82c575);
let p3 = f64::from_bits(0x3fac6b08d704a0a6);
let p4 = f64::from_bits(0x3f83b2ab88f70400);
let p5 = f64::from_bits(0x3f55d88003875c74);
// double_t r, t, z;
// uint32_t ix, i0;
// union {double f; uint64_t i;} u = {x};
// union {uint32_t u; int32_t i;} k;
let x1p1023 = f64::from_bits(0x7fe0000000000000);
let x1p52 = f64::from_bits(0x4330000000000000);
let _0x1p_149 = f64::from_bits(0xb6a0000000000000);
/* Filter out exceptional cases. */
let ui = f64::to_bits(x);
let ix = ui >> 32 & 0x7fffffff;
if ix >= 0x408ff000 {
/* |x| >= 1022 or nan */
if ix >= 0x40900000 && ui >> 63 == 0 {
/* x >= 1024 or nan */
/* overflow */
x *= x1p1023;
return x;
}
if ix >= 0x7ff00000 {
/* -inf or -nan */
return -1.0 / x;
}
if ui >> 63 != 0 {
/* x <= -1022 */
/* underflow */
if x <= -1075.0 || x - x1p52 + x1p52 != x {
force_eval!((_0x1p_149 / x) as f32);
}
if x <= -1075.0 {
return 0.0;
}
}
} else if ix < 0x3c900000 {
/* |x| < 0x1p-54 */
return 1.0 + x;
}
/* Reduce x, computing z, i0, and k. */
let ui = f64::to_bits(x + redux);
let mut i0 = ui as u32;
i0 = i0.wrapping_add(TBLSIZE as u32 / 2);
let ku = i0 / TBLSIZE as u32 * TBLSIZE as u32;
let ki = div!(ku as i32, TBLSIZE as i32);
i0 %= TBLSIZE as u32;
let uf = f64::from_bits(ui) - redux;
let mut z = x - uf;
/* Compute r = exp2(y) = exp2t[i0] * p(z - eps[i]). */
let t = f64::from_bits(i!(TBL, 2 * i0 as usize)); /* exp2t[i0] */
z -= f64::from_bits(i!(TBL, 2 * i0 as usize + 1)); /* eps[i0] */
let r = t + t * z * (p1 + z * (p2 + z * (p3 + z * (p4 + z * p5))));
scalbn(r, ki)
}
#[test]
fn i0_wrap_test() {
let x = -3.0 / 256.0;
assert_eq!(exp2(x), f64::from_bits(0x3fefbdba3692d514));
}

View File

@ -0,0 +1,135 @@
// origin: FreeBSD /usr/src/lib/msun/src/s_exp2f.c
//-
// Copyright (c) 2005 David Schultz <das@FreeBSD.ORG>
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
// 1. Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// 2. Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
// ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
// OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
// HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
// OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
// SUCH DAMAGE.
const TBLSIZE: usize = 16;
static EXP2FT: [u64; TBLSIZE] = [
0x3fe6a09e667f3bcd,
0x3fe7a11473eb0187,
0x3fe8ace5422aa0db,
0x3fe9c49182a3f090,
0x3feae89f995ad3ad,
0x3fec199bdd85529c,
0x3fed5818dcfba487,
0x3feea4afa2a490da,
0x3ff0000000000000,
0x3ff0b5586cf9890f,
0x3ff172b83c7d517b,
0x3ff2387a6e756238,
0x3ff306fe0a31b715,
0x3ff3dea64c123422,
0x3ff4bfdad5362a27,
0x3ff5ab07dd485429,
];
// exp2f(x): compute the base 2 exponential of x
//
// Accuracy: Peak error < 0.501 ulp; location of peak: -0.030110927.
//
// Method: (equally-spaced tables)
//
// Reduce x:
// x = k + y, for integer k and |y| <= 1/2.
// Thus we have exp2f(x) = 2**k * exp2(y).
//
// Reduce y:
// y = i/TBLSIZE + z for integer i near y * TBLSIZE.
// Thus we have exp2(y) = exp2(i/TBLSIZE) * exp2(z),
// with |z| <= 2**-(TBLSIZE+1).
//
// We compute exp2(i/TBLSIZE) via table lookup and exp2(z) via a
// degree-4 minimax polynomial with maximum error under 1.4 * 2**-33.
// Using double precision for everything except the reduction makes
// roundoff error insignificant and simplifies the scaling step.
//
// This method is due to Tang, but I do not use his suggested parameters:
//
// Tang, P. Table-driven Implementation of the Exponential Function
// in IEEE Floating-Point Arithmetic. TOMS 15(2), 144-157 (1989).
/// Exponential, base 2 (f32)
///
/// Calculate `2^x`, that is, 2 raised to the power `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn exp2f(mut x: f32) -> f32 {
let redux = f32::from_bits(0x4b400000) / TBLSIZE as f32;
let p1 = f32::from_bits(0x3f317218);
let p2 = f32::from_bits(0x3e75fdf0);
let p3 = f32::from_bits(0x3d6359a4);
let p4 = f32::from_bits(0x3c1d964e);
// double_t t, r, z;
// uint32_t ix, i0, k;
let x1p127 = f32::from_bits(0x7f000000);
/* Filter out exceptional cases. */
let ui = f32::to_bits(x);
let ix = ui & 0x7fffffff;
if ix > 0x42fc0000 {
/* |x| > 126 */
if ix > 0x7f800000 {
/* NaN */
return x;
}
if ui >= 0x43000000 && ui < 0x80000000 {
/* x >= 128 */
x *= x1p127;
return x;
}
if ui >= 0x80000000 {
/* x < -126 */
if ui >= 0xc3160000 || (ui & 0x0000ffff != 0) {
force_eval!(f32::from_bits(0x80000001) / x);
}
if ui >= 0xc3160000 {
/* x <= -150 */
return 0.0;
}
}
} else if ix <= 0x33000000 {
/* |x| <= 0x1p-25 */
return 1.0 + x;
}
/* Reduce x, computing z, i0, and k. */
let ui = f32::to_bits(x + redux);
let mut i0 = ui;
i0 += TBLSIZE as u32 / 2;
let k = i0 / TBLSIZE as u32;
let ukf = f64::from_bits(((0x3ff + k) as u64) << 52);
i0 &= TBLSIZE as u32 - 1;
let mut uf = f32::from_bits(ui);
uf -= redux;
let z: f64 = (x - uf) as f64;
/* Compute r = exp2(y) = exp2ft[i0] * p(z). */
let r: f64 = f64::from_bits(i!(EXP2FT, i0 as usize));
let t: f64 = r as f64 * z;
let r: f64 = r + t * (p1 as f64 + z * p2 as f64) + t * (z * z) * (p3 as f64 + z * p4 as f64);
/* Scale by 2**k */
(r * ukf) as f32
}

View File

@ -0,0 +1,97 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_expf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::scalbnf;
const HALF: [f32; 2] = [0.5, -0.5];
const LN2_HI: f32 = 6.9314575195e-01; /* 0x3f317200 */
const LN2_LO: f32 = 1.4286067653e-06; /* 0x35bfbe8e */
const INV_LN2: f32 = 1.4426950216e+00; /* 0x3fb8aa3b */
/*
* Domain [-0.34568, 0.34568], range ~[-4.278e-9, 4.447e-9]:
* |x*(exp(x)+1)/(exp(x)-1) - p(x)| < 2**-27.74
*/
const P1: f32 = 1.6666625440e-1; /* 0xaaaa8f.0p-26 */
const P2: f32 = -2.7667332906e-3; /* -0xb55215.0p-32 */
/// Exponential, base *e* (f32)
///
/// Calculate the exponential of `x`, that is, *e* raised to the power `x`
/// (where *e* is the base of the natural system of logarithms, approximately 2.71828).
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn expf(mut x: f32) -> f32 {
let x1p127 = f32::from_bits(0x7f000000); // 0x1p127f === 2 ^ 127
let x1p_126 = f32::from_bits(0x800000); // 0x1p-126f === 2 ^ -126 /*original 0x1p-149f ??????????? */
let mut hx = x.to_bits();
let sign = (hx >> 31) as i32; /* sign bit of x */
let signb: bool = sign != 0;
hx &= 0x7fffffff; /* high word of |x| */
/* special cases */
if hx >= 0x42aeac50 {
/* if |x| >= -87.33655f or NaN */
if hx > 0x7f800000 {
/* NaN */
return x;
}
if (hx >= 0x42b17218) && (!signb) {
/* x >= 88.722839f */
/* overflow */
x *= x1p127;
return x;
}
if signb {
/* underflow */
force_eval!(-x1p_126 / x);
if hx >= 0x42cff1b5 {
/* x <= -103.972084f */
return 0.;
}
}
}
/* argument reduction */
let k: i32;
let hi: f32;
let lo: f32;
if hx > 0x3eb17218 {
/* if |x| > 0.5 ln2 */
if hx > 0x3f851592 {
/* if |x| > 1.5 ln2 */
k = (INV_LN2 * x + i!(HALF, sign as usize)) as i32;
} else {
k = 1 - sign - sign;
}
let kf = k as f32;
hi = x - kf * LN2_HI; /* k*ln2hi is exact here */
lo = kf * LN2_LO;
x = hi - lo;
} else if hx > 0x39000000 {
/* |x| > 2**-14 */
k = 0;
hi = x;
lo = 0.;
} else {
/* raise inexact */
force_eval!(x1p127 + x);
return 1. + x;
}
/* x is now in primary range */
let xx = x * x;
let c = x - xx * (P1 + xx * P2);
let y = 1. + (x * c / (2. - c) - lo + hi);
if k == 0 { y } else { scalbnf(y, k) }
}

View File

@ -0,0 +1,144 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_expm1.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use core::f64;
const O_THRESHOLD: f64 = 7.09782712893383973096e+02; /* 0x40862E42, 0xFEFA39EF */
const LN2_HI: f64 = 6.93147180369123816490e-01; /* 0x3fe62e42, 0xfee00000 */
const LN2_LO: f64 = 1.90821492927058770002e-10; /* 0x3dea39ef, 0x35793c76 */
const INVLN2: f64 = 1.44269504088896338700e+00; /* 0x3ff71547, 0x652b82fe */
/* Scaled Q's: Qn_here = 2**n * Qn_above, for R(2*z) where z = hxs = x*x/2: */
const Q1: f64 = -3.33333333333331316428e-02; /* BFA11111 111110F4 */
const Q2: f64 = 1.58730158725481460165e-03; /* 3F5A01A0 19FE5585 */
const Q3: f64 = -7.93650757867487942473e-05; /* BF14CE19 9EAADBB7 */
const Q4: f64 = 4.00821782732936239552e-06; /* 3ED0CFCA 86E65239 */
const Q5: f64 = -2.01099218183624371326e-07; /* BE8AFDB7 6E09C32D */
/// Exponential, base *e*, of x-1 (f64)
///
/// Calculates the exponential of `x` and subtract 1, that is, *e* raised
/// to the power `x` minus 1 (where *e* is the base of the natural
/// system of logarithms, approximately 2.71828).
/// The result is accurate even for small values of `x`,
/// where using `exp(x)-1` would lose many significant digits.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn expm1(mut x: f64) -> f64 {
let hi: f64;
let lo: f64;
let k: i32;
let c: f64;
let mut t: f64;
let mut y: f64;
let mut ui = x.to_bits();
let hx = ((ui >> 32) & 0x7fffffff) as u32;
let sign = (ui >> 63) as i32;
/* filter out huge and non-finite argument */
if hx >= 0x4043687A {
/* if |x|>=56*ln2 */
if x.is_nan() {
return x;
}
if sign != 0 {
return -1.0;
}
if x > O_THRESHOLD {
x *= f64::from_bits(0x7fe0000000000000);
return x;
}
}
/* argument reduction */
if hx > 0x3fd62e42 {
/* if |x| > 0.5 ln2 */
if hx < 0x3FF0A2B2 {
/* and |x| < 1.5 ln2 */
if sign == 0 {
hi = x - LN2_HI;
lo = LN2_LO;
k = 1;
} else {
hi = x + LN2_HI;
lo = -LN2_LO;
k = -1;
}
} else {
k = (INVLN2 * x + if sign != 0 { -0.5 } else { 0.5 }) as i32;
t = k as f64;
hi = x - t * LN2_HI; /* t*ln2_hi is exact here */
lo = t * LN2_LO;
}
x = hi - lo;
c = (hi - x) - lo;
} else if hx < 0x3c900000 {
/* |x| < 2**-54, return x */
if hx < 0x00100000 {
force_eval!(x);
}
return x;
} else {
c = 0.0;
k = 0;
}
/* x is now in primary range */
let hfx = 0.5 * x;
let hxs = x * hfx;
let r1 = 1.0 + hxs * (Q1 + hxs * (Q2 + hxs * (Q3 + hxs * (Q4 + hxs * Q5))));
t = 3.0 - r1 * hfx;
let mut e = hxs * ((r1 - t) / (6.0 - x * t));
if k == 0 {
/* c is 0 */
return x - (x * e - hxs);
}
e = x * (e - c) - c;
e -= hxs;
/* exp(x) ~ 2^k (x_reduced - e + 1) */
if k == -1 {
return 0.5 * (x - e) - 0.5;
}
if k == 1 {
if x < -0.25 {
return -2.0 * (e - (x + 0.5));
}
return 1.0 + 2.0 * (x - e);
}
ui = ((0x3ff + k) as u64) << 52; /* 2^k */
let twopk = f64::from_bits(ui);
if k < 0 || k > 56 {
/* suffice to return exp(x)-1 */
y = x - e + 1.0;
if k == 1024 {
y = y * 2.0 * f64::from_bits(0x7fe0000000000000);
} else {
y = y * twopk;
}
return y - 1.0;
}
ui = ((0x3ff - k) as u64) << 52; /* 2^-k */
let uf = f64::from_bits(ui);
if k < 20 {
y = (x - e + (1.0 - uf)) * twopk;
} else {
y = (x - (e + uf) + 1.0) * twopk;
}
y
}
#[cfg(test)]
mod tests {
#[test]
fn sanity_check() {
assert_eq!(super::expm1(1.1), 2.0041660239464334);
}
}

View File

@ -0,0 +1,130 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_expm1f.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
const O_THRESHOLD: f32 = 8.8721679688e+01; /* 0x42b17180 */
const LN2_HI: f32 = 6.9313812256e-01; /* 0x3f317180 */
const LN2_LO: f32 = 9.0580006145e-06; /* 0x3717f7d1 */
const INV_LN2: f32 = 1.4426950216e+00; /* 0x3fb8aa3b */
/*
* Domain [-0.34568, 0.34568], range ~[-6.694e-10, 6.696e-10]:
* |6 / x * (1 + 2 * (1 / (exp(x) - 1) - 1 / x)) - q(x)| < 2**-30.04
* Scaled coefficients: Qn_here = 2**n * Qn_for_q (see s_expm1.c):
*/
const Q1: f32 = -3.3333212137e-2; /* -0x888868.0p-28 */
const Q2: f32 = 1.5807170421e-3; /* 0xcf3010.0p-33 */
/// Exponential, base *e*, of x-1 (f32)
///
/// Calculates the exponential of `x` and subtract 1, that is, *e* raised
/// to the power `x` minus 1 (where *e* is the base of the natural
/// system of logarithms, approximately 2.71828).
/// The result is accurate even for small values of `x`,
/// where using `exp(x)-1` would lose many significant digits.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn expm1f(mut x: f32) -> f32 {
let x1p127 = f32::from_bits(0x7f000000); // 0x1p127f === 2 ^ 127
let mut hx = x.to_bits();
let sign = (hx >> 31) != 0;
hx &= 0x7fffffff;
/* filter out huge and non-finite argument */
if hx >= 0x4195b844 {
/* if |x|>=27*ln2 */
if hx > 0x7f800000 {
/* NaN */
return x;
}
if sign {
return -1.;
}
if x > O_THRESHOLD {
x *= x1p127;
return x;
}
}
let k: i32;
let hi: f32;
let lo: f32;
let mut c = 0f32;
/* argument reduction */
if hx > 0x3eb17218 {
/* if |x| > 0.5 ln2 */
if hx < 0x3F851592 {
/* and |x| < 1.5 ln2 */
if !sign {
hi = x - LN2_HI;
lo = LN2_LO;
k = 1;
} else {
hi = x + LN2_HI;
lo = -LN2_LO;
k = -1;
}
} else {
k = (INV_LN2 * x + (if sign { -0.5 } else { 0.5 })) as i32;
let t = k as f32;
hi = x - t * LN2_HI; /* t*ln2_hi is exact here */
lo = t * LN2_LO;
}
x = hi - lo;
c = (hi - x) - lo;
} else if hx < 0x33000000 {
/* when |x|<2**-25, return x */
if hx < 0x00800000 {
force_eval!(x * x);
}
return x;
} else {
k = 0;
}
/* x is now in primary range */
let hfx = 0.5 * x;
let hxs = x * hfx;
let r1 = 1. + hxs * (Q1 + hxs * Q2);
let t = 3. - r1 * hfx;
let mut e = hxs * ((r1 - t) / (6. - x * t));
if k == 0 {
/* c is 0 */
return x - (x * e - hxs);
}
e = x * (e - c) - c;
e -= hxs;
/* exp(x) ~ 2^k (x_reduced - e + 1) */
if k == -1 {
return 0.5 * (x - e) - 0.5;
}
if k == 1 {
if x < -0.25 {
return -2. * (e - (x + 0.5));
}
return 1. + 2. * (x - e);
}
let twopk = f32::from_bits(((0x7f + k) << 23) as u32); /* 2^k */
if (k < 0) || (k > 56) {
/* suffice to return exp(x)-1 */
let mut y = x - e + 1.;
if k == 128 {
y = y * 2. * x1p127;
} else {
y = y * twopk;
}
return y - 1.;
}
let uf = f32::from_bits(((0x7f - k) << 23) as u32); /* 2^-k */
if k < 23 { (x - e + (1. - uf)) * twopk } else { (x - (e + uf) + 1.) * twopk }
}

View File

@ -0,0 +1,14 @@
use super::{combine_words, exp};
/* exp(x)/2 for x >= log(DBL_MAX), slightly better than 0.5*exp(x/2)*exp(x/2) */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn expo2(x: f64) -> f64 {
/* k is such that k*ln2 has minimal relative error and x - kln2 > log(DBL_MIN) */
const K: i32 = 2043;
let kln2 = f64::from_bits(0x40962066151add8b);
/* note that k is odd and scale*scale overflows */
let scale = combine_words(((0x3ff + K / 2) as u32) << 20, 0);
/* exp(x - k ln2) * 2**(k-1) */
exp(x - kln2) * scale * scale
}

View File

@ -0,0 +1,41 @@
use core::u64;
/// Absolute value (magnitude) (f64)
/// Calculates the absolute value (magnitude) of the argument `x`,
/// by direct manipulation of the bit representation of `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fabs(x: f64) -> f64 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f64.abs` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::fabsf64(x) }
}
}
f64::from_bits(x.to_bits() & (u64::MAX / 2))
}
#[cfg(test)]
mod tests {
use super::*;
use core::f64::*;
#[test]
fn sanity_check() {
assert_eq!(fabs(-1.0), 1.0);
assert_eq!(fabs(2.8), 2.8);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/fabs
#[test]
fn spec_tests() {
assert!(fabs(NAN).is_nan());
for f in [0.0, -0.0].iter().copied() {
assert_eq!(fabs(f), 0.0);
}
for f in [INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(fabs(f), INFINITY);
}
}
}

View File

@ -0,0 +1,41 @@
/// Absolute value (magnitude) (f32)
/// Calculates the absolute value (magnitude) of the argument `x`,
/// by direct manipulation of the bit representation of `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fabsf(x: f32) -> f32 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f32.abs` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::fabsf32(x) }
}
}
f32::from_bits(x.to_bits() & 0x7fffffff)
}
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
#[cfg(not(target_arch = "powerpc64"))]
#[cfg(test)]
mod tests {
use super::*;
use core::f32::*;
#[test]
fn sanity_check() {
assert_eq!(fabsf(-1.0), 1.0);
assert_eq!(fabsf(2.8), 2.8);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/fabs
#[test]
fn spec_tests() {
assert!(fabsf(NAN).is_nan());
for f in [0.0, -0.0].iter().copied() {
assert_eq!(fabsf(f), 0.0);
}
for f in [INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(fabsf(f), INFINITY);
}
}
}

View File

@ -0,0 +1,22 @@
use core::f64;
/// Positive difference (f64)
///
/// Determines the positive difference between arguments, returning:
/// * x - y if x > y, or
/// * +0 if x <= y, or
/// * NAN if either argument is NAN.
///
/// A range error may occur.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fdim(x: f64, y: f64) -> f64 {
if x.is_nan() {
x
} else if y.is_nan() {
y
} else if x > y {
x - y
} else {
0.0
}
}

View File

@ -0,0 +1,22 @@
use core::f32;
/// Positive difference (f32)
///
/// Determines the positive difference between arguments, returning:
/// * x - y if x > y, or
/// * +0 if x <= y, or
/// * NAN if either argument is NAN.
///
/// A range error may occur.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fdimf(x: f32, y: f32) -> f32 {
if x.is_nan() {
x
} else if y.is_nan() {
y
} else if x > y {
x - y
} else {
0.0
}
}

View File

@ -0,0 +1,27 @@
// src: musl/src/fenv/fenv.c
/* Dummy functions for archs lacking fenv implementation */
pub(crate) const FE_UNDERFLOW: i32 = 0;
pub(crate) const FE_INEXACT: i32 = 0;
pub(crate) const FE_TONEAREST: i32 = 0;
#[inline]
pub(crate) fn feclearexcept(_mask: i32) -> i32 {
0
}
#[inline]
pub(crate) fn feraiseexcept(_mask: i32) -> i32 {
0
}
#[inline]
pub(crate) fn fetestexcept(_mask: i32) -> i32 {
0
}
#[inline]
pub(crate) fn fegetround() -> i32 {
FE_TONEAREST
}

View File

@ -0,0 +1,73 @@
#![allow(unreachable_code)]
use core::f64;
const TOINT: f64 = 1. / f64::EPSILON;
/// Floor (f64)
///
/// Finds the nearest integer less than or equal to `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn floor(x: f64) -> f64 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f64.floor` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::floorf64(x) }
}
}
#[cfg(all(target_arch = "x86", not(target_feature = "sse2")))]
{
//use an alternative implementation on x86, because the
//main implementation fails with the x87 FPU used by
//debian i386, probablly due to excess precision issues.
//basic implementation taken from https://github.com/rust-lang/libm/issues/219
use super::fabs;
if fabs(x).to_bits() < 4503599627370496.0_f64.to_bits() {
let truncated = x as i64 as f64;
if truncated > x {
return truncated - 1.0;
} else {
return truncated;
}
} else {
return x;
}
}
let ui = x.to_bits();
let e = ((ui >> 52) & 0x7ff) as i32;
if (e >= 0x3ff + 52) || (x == 0.) {
return x;
}
/* y = int(x) - x, where int(x) is an integer neighbor of x */
let y = if (ui >> 63) != 0 { x - TOINT + TOINT - x } else { x + TOINT - TOINT - x };
/* special case because of non-nearest rounding modes */
if e < 0x3ff {
force_eval!(y);
return if (ui >> 63) != 0 { -1. } else { 0. };
}
if y > 0. { x + y - 1. } else { x + y }
}
#[cfg(test)]
mod tests {
use super::*;
use core::f64::*;
#[test]
fn sanity_check() {
assert_eq!(floor(1.1), 1.0);
assert_eq!(floor(2.9), 2.0);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/floor
#[test]
fn spec_tests() {
// Not Asserted: that the current rounding mode has no effect.
assert!(floor(NAN).is_nan());
for f in [0.0, -0.0, INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(floor(f), f);
}
}
}

View File

@ -0,0 +1,66 @@
use core::f32;
/// Floor (f32)
///
/// Finds the nearest integer less than or equal to `x`.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn floorf(x: f32) -> f32 {
// On wasm32 we know that LLVM's intrinsic will compile to an optimized
// `f32.floor` native instruction, so we can leverage this for both code size
// and speed.
llvm_intrinsically_optimized! {
#[cfg(target_arch = "wasm32")] {
return unsafe { ::core::intrinsics::floorf32(x) }
}
}
let mut ui = x.to_bits();
let e = (((ui >> 23) as i32) & 0xff) - 0x7f;
if e >= 23 {
return x;
}
if e >= 0 {
let m: u32 = 0x007fffff >> e;
if (ui & m) == 0 {
return x;
}
force_eval!(x + f32::from_bits(0x7b800000));
if ui >> 31 != 0 {
ui += m;
}
ui &= !m;
} else {
force_eval!(x + f32::from_bits(0x7b800000));
if ui >> 31 == 0 {
ui = 0;
} else if ui << 1 != 0 {
return -1.0;
}
}
f32::from_bits(ui)
}
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
#[cfg(not(target_arch = "powerpc64"))]
#[cfg(test)]
mod tests {
use super::*;
use core::f32::*;
#[test]
fn sanity_check() {
assert_eq!(floorf(0.5), 0.0);
assert_eq!(floorf(1.1), 1.0);
assert_eq!(floorf(2.9), 2.0);
}
/// The spec: https://en.cppreference.com/w/cpp/numeric/math/floor
#[test]
fn spec_tests() {
// Not Asserted: that the current rounding mode has no effect.
assert!(floorf(NAN).is_nan());
for f in [0.0, -0.0, INFINITY, NEG_INFINITY].iter().copied() {
assert_eq!(floorf(f), f);
}
}
}

View File

@ -0,0 +1,226 @@
use core::{f32, f64};
use super::scalbn;
const ZEROINFNAN: i32 = 0x7ff - 0x3ff - 52 - 1;
struct Num {
m: u64,
e: i32,
sign: i32,
}
fn normalize(x: f64) -> Num {
let x1p63: f64 = f64::from_bits(0x43e0000000000000); // 0x1p63 === 2 ^ 63
let mut ix: u64 = x.to_bits();
let mut e: i32 = (ix >> 52) as i32;
let sign: i32 = e & 0x800;
e &= 0x7ff;
if e == 0 {
ix = (x * x1p63).to_bits();
e = (ix >> 52) as i32 & 0x7ff;
e = if e != 0 { e - 63 } else { 0x800 };
}
ix &= (1 << 52) - 1;
ix |= 1 << 52;
ix <<= 1;
e -= 0x3ff + 52 + 1;
Num { m: ix, e, sign }
}
#[inline]
fn mul(x: u64, y: u64) -> (u64, u64) {
let t = (x as u128).wrapping_mul(y as u128);
((t >> 64) as u64, t as u64)
}
/// Floating multiply add (f64)
///
/// Computes `(x*y)+z`, rounded as one ternary operation:
/// Computes the value (as if) to infinite precision and rounds once to the result format,
/// according to the rounding mode characterized by the value of FLT_ROUNDS.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fma(x: f64, y: f64, z: f64) -> f64 {
let x1p63: f64 = f64::from_bits(0x43e0000000000000); // 0x1p63 === 2 ^ 63
let x0_ffffff8p_63 = f64::from_bits(0x3bfffffff0000000); // 0x0.ffffff8p-63
/* normalize so top 10bits and last bit are 0 */
let nx = normalize(x);
let ny = normalize(y);
let nz = normalize(z);
if nx.e >= ZEROINFNAN || ny.e >= ZEROINFNAN {
return x * y + z;
}
if nz.e >= ZEROINFNAN {
if nz.e > ZEROINFNAN {
/* z==0 */
return x * y + z;
}
return z;
}
/* mul: r = x*y */
let zhi: u64;
let zlo: u64;
let (mut rhi, mut rlo) = mul(nx.m, ny.m);
/* either top 20 or 21 bits of rhi and last 2 bits of rlo are 0 */
/* align exponents */
let mut e: i32 = nx.e + ny.e;
let mut d: i32 = nz.e - e;
/* shift bits z<<=kz, r>>=kr, so kz+kr == d, set e = e+kr (== ez-kz) */
if d > 0 {
if d < 64 {
zlo = nz.m << d;
zhi = nz.m >> (64 - d);
} else {
zlo = 0;
zhi = nz.m;
e = nz.e - 64;
d -= 64;
if d == 0 {
} else if d < 64 {
rlo = rhi << (64 - d) | rlo >> d | ((rlo << (64 - d)) != 0) as u64;
rhi = rhi >> d;
} else {
rlo = 1;
rhi = 0;
}
}
} else {
zhi = 0;
d = -d;
if d == 0 {
zlo = nz.m;
} else if d < 64 {
zlo = nz.m >> d | ((nz.m << (64 - d)) != 0) as u64;
} else {
zlo = 1;
}
}
/* add */
let mut sign: i32 = nx.sign ^ ny.sign;
let samesign: bool = (sign ^ nz.sign) == 0;
let mut nonzero: i32 = 1;
if samesign {
/* r += z */
rlo = rlo.wrapping_add(zlo);
rhi += zhi + (rlo < zlo) as u64;
} else {
/* r -= z */
let (res, borrow) = rlo.overflowing_sub(zlo);
rlo = res;
rhi = rhi.wrapping_sub(zhi.wrapping_add(borrow as u64));
if (rhi >> 63) != 0 {
rlo = (rlo as i64).wrapping_neg() as u64;
rhi = (rhi as i64).wrapping_neg() as u64 - (rlo != 0) as u64;
sign = (sign == 0) as i32;
}
nonzero = (rhi != 0) as i32;
}
/* set rhi to top 63bit of the result (last bit is sticky) */
if nonzero != 0 {
e += 64;
d = rhi.leading_zeros() as i32 - 1;
/* note: d > 0 */
rhi = rhi << d | rlo >> (64 - d) | ((rlo << d) != 0) as u64;
} else if rlo != 0 {
d = rlo.leading_zeros() as i32 - 1;
if d < 0 {
rhi = rlo >> 1 | (rlo & 1);
} else {
rhi = rlo << d;
}
} else {
/* exact +-0 */
return x * y + z;
}
e -= d;
/* convert to double */
let mut i: i64 = rhi as i64; /* i is in [1<<62,(1<<63)-1] */
if sign != 0 {
i = -i;
}
let mut r: f64 = i as f64; /* |r| is in [0x1p62,0x1p63] */
if e < -1022 - 62 {
/* result is subnormal before rounding */
if e == -1022 - 63 {
let mut c: f64 = x1p63;
if sign != 0 {
c = -c;
}
if r == c {
/* min normal after rounding, underflow depends
on arch behaviour which can be imitated by
a double to float conversion */
let fltmin: f32 = (x0_ffffff8p_63 * f32::MIN_POSITIVE as f64 * r) as f32;
return f64::MIN_POSITIVE / f32::MIN_POSITIVE as f64 * fltmin as f64;
}
/* one bit is lost when scaled, add another top bit to
only round once at conversion if it is inexact */
if (rhi << 53) != 0 {
i = (rhi >> 1 | (rhi & 1) | 1 << 62) as i64;
if sign != 0 {
i = -i;
}
r = i as f64;
r = 2. * r - c; /* remove top bit */
/* raise underflow portably, such that it
cannot be optimized away */
{
let tiny: f64 = f64::MIN_POSITIVE / f32::MIN_POSITIVE as f64 * r;
r += (tiny * tiny) * (r - r);
}
}
} else {
/* only round once when scaled */
d = 10;
i = ((rhi >> d | ((rhi << (64 - d)) != 0) as u64) << d) as i64;
if sign != 0 {
i = -i;
}
r = i as f64;
}
}
scalbn(r, e)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn fma_segfault() {
// These two inputs cause fma to segfault on release due to overflow:
assert_eq!(
fma(
-0.0000000000000002220446049250313,
-0.0000000000000002220446049250313,
-0.0000000000000002220446049250313
),
-0.00000000000000022204460492503126,
);
let result = fma(-0.992, -0.992, -0.992);
//force rounding to storage format on x87 to prevent superious errors.
#[cfg(all(target_arch = "x86", not(target_feature = "sse2")))]
let result = force_eval!(result);
assert_eq!(result, -0.007936000000000007,);
}
#[test]
fn fma_sbb() {
assert_eq!(fma(-(1.0 - f64::EPSILON), f64::MIN, f64::MIN), -3991680619069439e277);
}
#[test]
fn fma_underflow() {
assert_eq!(fma(1.1102230246251565e-16, -9.812526705433188e-305, 1.0894e-320), 0.0,);
}
}

View File

@ -0,0 +1,113 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_fmaf.c */
/*-
* Copyright (c) 2005-2011 David Schultz <das@FreeBSD.ORG>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
use core::f32;
use core::ptr::read_volatile;
use super::fenv::{
feclearexcept, fegetround, feraiseexcept, fetestexcept, FE_INEXACT, FE_TONEAREST, FE_UNDERFLOW,
};
/*
* Fused multiply-add: Compute x * y + z with a single rounding error.
*
* A double has more than twice as much precision than a float, so
* direct double-precision arithmetic suffices, except where double
* rounding occurs.
*/
/// Floating multiply add (f32)
///
/// Computes `(x*y)+z`, rounded as one ternary operation:
/// Computes the value (as if) to infinite precision and rounds once to the result format,
/// according to the rounding mode characterized by the value of FLT_ROUNDS.
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmaf(x: f32, y: f32, mut z: f32) -> f32 {
let xy: f64;
let mut result: f64;
let mut ui: u64;
let e: i32;
xy = x as f64 * y as f64;
result = xy + z as f64;
ui = result.to_bits();
e = (ui >> 52) as i32 & 0x7ff;
/* Common case: The double precision result is fine. */
if (
/* not a halfway case */
ui & 0x1fffffff) != 0x10000000 ||
/* NaN */
e == 0x7ff ||
/* exact */
(result - xy == z as f64 && result - z as f64 == xy) ||
/* not round-to-nearest */
fegetround() != FE_TONEAREST
{
/*
underflow may not be raised correctly, example:
fmaf(0x1p-120f, 0x1p-120f, 0x1p-149f)
*/
if e < 0x3ff - 126 && e >= 0x3ff - 149 && fetestexcept(FE_INEXACT) != 0 {
feclearexcept(FE_INEXACT);
// prevent `xy + vz` from being CSE'd with `xy + z` above
let vz: f32 = unsafe { read_volatile(&z) };
result = xy + vz as f64;
if fetestexcept(FE_INEXACT) != 0 {
feraiseexcept(FE_UNDERFLOW);
} else {
feraiseexcept(FE_INEXACT);
}
}
z = result as f32;
return z;
}
/*
* If result is inexact, and exactly halfway between two float values,
* we need to adjust the low-order bit in the direction of the error.
*/
let neg = ui >> 63 != 0;
let err = if neg == (z as f64 > xy) { xy - result + z as f64 } else { z as f64 - result + xy };
if neg == (err < 0.0) {
ui += 1;
} else {
ui -= 1;
}
f64::from_bits(ui) as f32
}
#[cfg(test)]
mod tests {
#[test]
fn issue_263() {
let a = f32::from_bits(1266679807);
let b = f32::from_bits(1300234242);
let c = f32::from_bits(1115553792);
let expected = f32::from_bits(1501560833);
assert_eq!(super::fmaf(a, b, c), expected);
}
}

View File

@ -0,0 +1,12 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmax(x: f64, y: f64) -> f64 {
// IEEE754 says: maxNum(x, y) is the canonicalized number y if x < y, x if y < x, the
// canonicalized number if one operand is a number and the other a quiet NaN. Otherwise it
// is either x or y, canonicalized (this means results might differ among implementations).
// When either x or y is a signalingNaN, then the result is according to 6.2.
//
// Since we do not support sNaN in Rust yet, we do not need to handle them.
// FIXME(nagisa): due to https://bugs.llvm.org/show_bug.cgi?id=33303 we canonicalize by
// multiplying by 1.0. Should switch to the `canonicalize` when it works.
(if x.is_nan() || x < y { y } else { x }) * 1.0
}

View File

@ -0,0 +1,12 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmaxf(x: f32, y: f32) -> f32 {
// IEEE754 says: maxNum(x, y) is the canonicalized number y if x < y, x if y < x, the
// canonicalized number if one operand is a number and the other a quiet NaN. Otherwise it
// is either x or y, canonicalized (this means results might differ among implementations).
// When either x or y is a signalingNaN, then the result is according to 6.2.
//
// Since we do not support sNaN in Rust yet, we do not need to handle them.
// FIXME(nagisa): due to https://bugs.llvm.org/show_bug.cgi?id=33303 we canonicalize by
// multiplying by 1.0. Should switch to the `canonicalize` when it works.
(if x.is_nan() || x < y { y } else { x }) * 1.0
}

View File

@ -0,0 +1,12 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmin(x: f64, y: f64) -> f64 {
// IEEE754 says: minNum(x, y) is the canonicalized number x if x < y, y if y < x, the
// canonicalized number if one operand is a number and the other a quiet NaN. Otherwise it
// is either x or y, canonicalized (this means results might differ among implementations).
// When either x or y is a signalingNaN, then the result is according to 6.2.
//
// Since we do not support sNaN in Rust yet, we do not need to handle them.
// FIXME(nagisa): due to https://bugs.llvm.org/show_bug.cgi?id=33303 we canonicalize by
// multiplying by 1.0. Should switch to the `canonicalize` when it works.
(if y.is_nan() || x < y { x } else { y }) * 1.0
}

View File

@ -0,0 +1,12 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fminf(x: f32, y: f32) -> f32 {
// IEEE754 says: minNum(x, y) is the canonicalized number x if x < y, y if y < x, the
// canonicalized number if one operand is a number and the other a quiet NaN. Otherwise it
// is either x or y, canonicalized (this means results might differ among implementations).
// When either x or y is a signalingNaN, then the result is according to 6.2.
//
// Since we do not support sNaN in Rust yet, we do not need to handle them.
// FIXME(nagisa): due to https://bugs.llvm.org/show_bug.cgi?id=33303 we canonicalize by
// multiplying by 1.0. Should switch to the `canonicalize` when it works.
(if y.is_nan() || x < y { x } else { y }) * 1.0
}

View File

@ -0,0 +1,80 @@
use core::u64;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmod(x: f64, y: f64) -> f64 {
let mut uxi = x.to_bits();
let mut uyi = y.to_bits();
let mut ex = (uxi >> 52 & 0x7ff) as i64;
let mut ey = (uyi >> 52 & 0x7ff) as i64;
let sx = uxi >> 63;
let mut i;
if uyi << 1 == 0 || y.is_nan() || ex == 0x7ff {
return (x * y) / (x * y);
}
if uxi << 1 <= uyi << 1 {
if uxi << 1 == uyi << 1 {
return 0.0 * x;
}
return x;
}
/* normalize x and y */
if ex == 0 {
i = uxi << 12;
while i >> 63 == 0 {
ex -= 1;
i <<= 1;
}
uxi <<= -ex + 1;
} else {
uxi &= u64::MAX >> 12;
uxi |= 1 << 52;
}
if ey == 0 {
i = uyi << 12;
while i >> 63 == 0 {
ey -= 1;
i <<= 1;
}
uyi <<= -ey + 1;
} else {
uyi &= u64::MAX >> 12;
uyi |= 1 << 52;
}
/* x mod y */
while ex > ey {
i = uxi.wrapping_sub(uyi);
if i >> 63 == 0 {
if i == 0 {
return 0.0 * x;
}
uxi = i;
}
uxi <<= 1;
ex -= 1;
}
i = uxi.wrapping_sub(uyi);
if i >> 63 == 0 {
if i == 0 {
return 0.0 * x;
}
uxi = i;
}
while uxi >> 52 == 0 {
uxi <<= 1;
ex -= 1;
}
/* scale result */
if ex > 0 {
uxi -= 1 << 52;
uxi |= (ex as u64) << 52;
} else {
uxi >>= -ex + 1;
}
uxi |= (sx as u64) << 63;
f64::from_bits(uxi)
}

View File

@ -0,0 +1,89 @@
use core::f32;
use core::u32;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn fmodf(x: f32, y: f32) -> f32 {
let mut uxi = x.to_bits();
let mut uyi = y.to_bits();
let mut ex = (uxi >> 23 & 0xff) as i32;
let mut ey = (uyi >> 23 & 0xff) as i32;
let sx = uxi & 0x80000000;
let mut i;
if uyi << 1 == 0 || y.is_nan() || ex == 0xff {
return (x * y) / (x * y);
}
if uxi << 1 <= uyi << 1 {
if uxi << 1 == uyi << 1 {
return 0.0 * x;
}
return x;
}
/* normalize x and y */
if ex == 0 {
i = uxi << 9;
while i >> 31 == 0 {
ex -= 1;
i <<= 1;
}
uxi <<= -ex + 1;
} else {
uxi &= u32::MAX >> 9;
uxi |= 1 << 23;
}
if ey == 0 {
i = uyi << 9;
while i >> 31 == 0 {
ey -= 1;
i <<= 1;
}
uyi <<= -ey + 1;
} else {
uyi &= u32::MAX >> 9;
uyi |= 1 << 23;
}
/* x mod y */
while ex > ey {
i = uxi.wrapping_sub(uyi);
if i >> 31 == 0 {
if i == 0 {
return 0.0 * x;
}
uxi = i;
}
uxi <<= 1;
ex -= 1;
}
i = uxi.wrapping_sub(uyi);
if i >> 31 == 0 {
if i == 0 {
return 0.0 * x;
}
uxi = i;
}
while uxi >> 23 == 0 {
uxi <<= 1;
ex -= 1;
}
/* scale result up */
if ex > 0 {
uxi -= 1 << 23;
uxi |= (ex as u32) << 23;
} else {
uxi >>= -ex + 1;
}
uxi |= sx;
f32::from_bits(uxi)
}

View File

@ -0,0 +1,20 @@
pub fn frexp(x: f64) -> (f64, i32) {
let mut y = x.to_bits();
let ee = ((y >> 52) & 0x7ff) as i32;
if ee == 0 {
if x != 0.0 {
let x1p64 = f64::from_bits(0x43f0000000000000);
let (x, e) = frexp(x * x1p64);
return (x, e - 64);
}
return (x, 0);
} else if ee == 0x7ff {
return (x, 0);
}
let e = ee - 0x3fe;
y &= 0x800fffffffffffff;
y |= 0x3fe0000000000000;
return (f64::from_bits(y), e);
}

View File

@ -0,0 +1,21 @@
pub fn frexpf(x: f32) -> (f32, i32) {
let mut y = x.to_bits();
let ee: i32 = ((y >> 23) & 0xff) as i32;
if ee == 0 {
if x != 0.0 {
let x1p64 = f32::from_bits(0x5f800000);
let (x, e) = frexpf(x * x1p64);
return (x, e - 64);
} else {
return (x, 0);
}
} else if ee == 0xff {
return (x, 0);
}
let e = ee - 0x7e;
y &= 0x807fffff;
y |= 0x3f000000;
(f32::from_bits(y), e)
}

View File

@ -0,0 +1,74 @@
use core::f64;
use super::sqrt;
const SPLIT: f64 = 134217728. + 1.; // 0x1p27 + 1 === (2 ^ 27) + 1
fn sq(x: f64) -> (f64, f64) {
let xh: f64;
let xl: f64;
let xc: f64;
xc = x * SPLIT;
xh = x - xc + xc;
xl = x - xh;
let hi = x * x;
let lo = xh * xh - hi + 2. * xh * xl + xl * xl;
(hi, lo)
}
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn hypot(mut x: f64, mut y: f64) -> f64 {
let x1p700 = f64::from_bits(0x6bb0000000000000); // 0x1p700 === 2 ^ 700
let x1p_700 = f64::from_bits(0x1430000000000000); // 0x1p-700 === 2 ^ -700
let mut uxi = x.to_bits();
let mut uyi = y.to_bits();
let uti;
let ex: i64;
let ey: i64;
let mut z: f64;
/* arrange |x| >= |y| */
uxi &= -1i64 as u64 >> 1;
uyi &= -1i64 as u64 >> 1;
if uxi < uyi {
uti = uxi;
uxi = uyi;
uyi = uti;
}
/* special cases */
ex = (uxi >> 52) as i64;
ey = (uyi >> 52) as i64;
x = f64::from_bits(uxi);
y = f64::from_bits(uyi);
/* note: hypot(inf,nan) == inf */
if ey == 0x7ff {
return y;
}
if ex == 0x7ff || uyi == 0 {
return x;
}
/* note: hypot(x,y) ~= x + y*y/x/2 with inexact for small y/x */
/* 64 difference is enough for ld80 double_t */
if ex - ey > 64 {
return x + y;
}
/* precise sqrt argument in nearest rounding mode without overflow */
/* xh*xh must not overflow and xl*xl must not underflow in sq */
z = 1.;
if ex > 0x3ff + 510 {
z = x1p700;
x *= x1p_700;
y *= x1p_700;
} else if ey < 0x3ff - 450 {
z = x1p_700;
x *= x1p700;
y *= x1p700;
}
let (hx, lx) = sq(x);
let (hy, ly) = sq(y);
z * sqrt(ly + lx + hy + hx)
}

View File

@ -0,0 +1,43 @@
use core::f32;
use super::sqrtf;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn hypotf(mut x: f32, mut y: f32) -> f32 {
let x1p90 = f32::from_bits(0x6c800000); // 0x1p90f === 2 ^ 90
let x1p_90 = f32::from_bits(0x12800000); // 0x1p-90f === 2 ^ -90
let mut uxi = x.to_bits();
let mut uyi = y.to_bits();
let uti;
let mut z: f32;
uxi &= -1i32 as u32 >> 1;
uyi &= -1i32 as u32 >> 1;
if uxi < uyi {
uti = uxi;
uxi = uyi;
uyi = uti;
}
x = f32::from_bits(uxi);
y = f32::from_bits(uyi);
if uyi == 0xff << 23 {
return y;
}
if uxi >= 0xff << 23 || uyi == 0 || uxi - uyi >= 25 << 23 {
return x + y;
}
z = 1.;
if uxi >= (0x7f + 60) << 23 {
z = x1p90;
x *= x1p_90;
y *= x1p_90;
} else if uyi < (0x7f - 60) << 23 {
z = x1p_90;
x *= x1p90;
y *= x1p90;
}
z * sqrtf((x as f64 * x as f64 + y as f64 * y as f64) as f32)
}

View File

@ -0,0 +1,28 @@
const FP_ILOGBNAN: i32 = -1 - 0x7fffffff;
const FP_ILOGB0: i32 = FP_ILOGBNAN;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ilogb(x: f64) -> i32 {
let mut i: u64 = x.to_bits();
let e = ((i >> 52) & 0x7ff) as i32;
if e == 0 {
i <<= 12;
if i == 0 {
force_eval!(0.0 / 0.0);
return FP_ILOGB0;
}
/* subnormal x */
let mut e = -0x3ff;
while (i >> 63) == 0 {
e -= 1;
i <<= 1;
}
e
} else if e == 0x7ff {
force_eval!(0.0 / 0.0);
if (i << 12) != 0 { FP_ILOGBNAN } else { i32::max_value() }
} else {
e - 0x3ff
}
}

View File

@ -0,0 +1,28 @@
const FP_ILOGBNAN: i32 = -1 - 0x7fffffff;
const FP_ILOGB0: i32 = FP_ILOGBNAN;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ilogbf(x: f32) -> i32 {
let mut i = x.to_bits();
let e = ((i >> 23) & 0xff) as i32;
if e == 0 {
i <<= 9;
if i == 0 {
force_eval!(0.0 / 0.0);
return FP_ILOGB0;
}
/* subnormal x */
let mut e = -0x7f;
while (i >> 31) == 0 {
e -= 1;
i <<= 1;
}
e
} else if e == 0xff {
force_eval!(0.0 / 0.0);
if (i << 9) != 0 { FP_ILOGBNAN } else { i32::max_value() }
} else {
e - 0x7f
}
}

View File

@ -0,0 +1,422 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_j0.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* j0(x), y0(x)
* Bessel function of the first and second kinds of order zero.
* Method -- j0(x):
* 1. For tiny x, we use j0(x) = 1 - x^2/4 + x^4/64 - ...
* 2. Reduce x to |x| since j0(x)=j0(-x), and
* for x in (0,2)
* j0(x) = 1-z/4+ z^2*R0/S0, where z = x*x;
* (precision: |j0-1+z/4-z^2R0/S0 |<2**-63.67 )
* for x in (2,inf)
* j0(x) = sqrt(2/(pi*x))*(p0(x)*cos(x0)-q0(x)*sin(x0))
* where x0 = x-pi/4. It is better to compute sin(x0),cos(x0)
* as follow:
* cos(x0) = cos(x)cos(pi/4)+sin(x)sin(pi/4)
* = 1/sqrt(2) * (cos(x) + sin(x))
* sin(x0) = sin(x)cos(pi/4)-cos(x)sin(pi/4)
* = 1/sqrt(2) * (sin(x) - cos(x))
* (To avoid cancellation, use
* sin(x) +- cos(x) = -cos(2x)/(sin(x) -+ cos(x))
* to compute the worse one.)
*
* 3 Special cases
* j0(nan)= nan
* j0(0) = 1
* j0(inf) = 0
*
* Method -- y0(x):
* 1. For x<2.
* Since
* y0(x) = 2/pi*(j0(x)*(ln(x/2)+Euler) + x^2/4 - ...)
* therefore y0(x)-2/pi*j0(x)*ln(x) is an even function.
* We use the following function to approximate y0,
* y0(x) = U(z)/V(z) + (2/pi)*(j0(x)*ln(x)), z= x^2
* where
* U(z) = u00 + u01*z + ... + u06*z^6
* V(z) = 1 + v01*z + ... + v04*z^4
* with absolute approximation error bounded by 2**-72.
* Note: For tiny x, U/V = u0 and j0(x)~1, hence
* y0(tiny) = u0 + (2/pi)*ln(tiny), (choose tiny<2**-27)
* 2. For x>=2.
* y0(x) = sqrt(2/(pi*x))*(p0(x)*cos(x0)+q0(x)*sin(x0))
* where x0 = x-pi/4. It is better to compute sin(x0),cos(x0)
* by the method mentioned above.
* 3. Special cases: y0(0)=-inf, y0(x<0)=NaN, y0(inf)=0.
*/
use super::{cos, fabs, get_high_word, get_low_word, log, sin, sqrt};
const INVSQRTPI: f64 = 5.64189583547756279280e-01; /* 0x3FE20DD7, 0x50429B6D */
const TPI: f64 = 6.36619772367581382433e-01; /* 0x3FE45F30, 0x6DC9C883 */
/* common method when |x|>=2 */
fn common(ix: u32, x: f64, y0: bool) -> f64 {
let s: f64;
let mut c: f64;
let mut ss: f64;
let mut cc: f64;
let z: f64;
/*
* j0(x) = sqrt(2/(pi*x))*(p0(x)*cos(x-pi/4)-q0(x)*sin(x-pi/4))
* y0(x) = sqrt(2/(pi*x))*(p0(x)*sin(x-pi/4)+q0(x)*cos(x-pi/4))
*
* sin(x-pi/4) = (sin(x) - cos(x))/sqrt(2)
* cos(x-pi/4) = (sin(x) + cos(x))/sqrt(2)
* sin(x) +- cos(x) = -cos(2x)/(sin(x) -+ cos(x))
*/
s = sin(x);
c = cos(x);
if y0 {
c = -c;
}
cc = s + c;
/* avoid overflow in 2*x, big ulp error when x>=0x1p1023 */
if ix < 0x7fe00000 {
ss = s - c;
z = -cos(2.0 * x);
if s * c < 0.0 {
cc = z / ss;
} else {
ss = z / cc;
}
if ix < 0x48000000 {
if y0 {
ss = -ss;
}
cc = pzero(x) * cc - qzero(x) * ss;
}
}
return INVSQRTPI * cc / sqrt(x);
}
/* R0/S0 on [0, 2.00] */
const R02: f64 = 1.56249999999999947958e-02; /* 0x3F8FFFFF, 0xFFFFFFFD */
const R03: f64 = -1.89979294238854721751e-04; /* 0xBF28E6A5, 0xB61AC6E9 */
const R04: f64 = 1.82954049532700665670e-06; /* 0x3EBEB1D1, 0x0C503919 */
const R05: f64 = -4.61832688532103189199e-09; /* 0xBE33D5E7, 0x73D63FCE */
const S01: f64 = 1.56191029464890010492e-02; /* 0x3F8FFCE8, 0x82C8C2A4 */
const S02: f64 = 1.16926784663337450260e-04; /* 0x3F1EA6D2, 0xDD57DBF4 */
const S03: f64 = 5.13546550207318111446e-07; /* 0x3EA13B54, 0xCE84D5A9 */
const S04: f64 = 1.16614003333790000205e-09; /* 0x3E1408BC, 0xF4745D8F */
pub fn j0(mut x: f64) -> f64 {
let z: f64;
let r: f64;
let s: f64;
let mut ix: u32;
ix = get_high_word(x);
ix &= 0x7fffffff;
/* j0(+-inf)=0, j0(nan)=nan */
if ix >= 0x7ff00000 {
return 1.0 / (x * x);
}
x = fabs(x);
if ix >= 0x40000000 {
/* |x| >= 2 */
/* large ulp error near zeros: 2.4, 5.52, 8.6537,.. */
return common(ix, x, false);
}
/* 1 - x*x/4 + x*x*R(x^2)/S(x^2) */
if ix >= 0x3f200000 {
/* |x| >= 2**-13 */
/* up to 4ulp error close to 2 */
z = x * x;
r = z * (R02 + z * (R03 + z * (R04 + z * R05)));
s = 1.0 + z * (S01 + z * (S02 + z * (S03 + z * S04)));
return (1.0 + x / 2.0) * (1.0 - x / 2.0) + z * (r / s);
}
/* 1 - x*x/4 */
/* prevent underflow */
/* inexact should be raised when x!=0, this is not done correctly */
if ix >= 0x38000000 {
/* |x| >= 2**-127 */
x = 0.25 * x * x;
}
return 1.0 - x;
}
const U00: f64 = -7.38042951086872317523e-02; /* 0xBFB2E4D6, 0x99CBD01F */
const U01: f64 = 1.76666452509181115538e-01; /* 0x3FC69D01, 0x9DE9E3FC */
const U02: f64 = -1.38185671945596898896e-02; /* 0xBF8C4CE8, 0xB16CFA97 */
const U03: f64 = 3.47453432093683650238e-04; /* 0x3F36C54D, 0x20B29B6B */
const U04: f64 = -3.81407053724364161125e-06; /* 0xBECFFEA7, 0x73D25CAD */
const U05: f64 = 1.95590137035022920206e-08; /* 0x3E550057, 0x3B4EABD4 */
const U06: f64 = -3.98205194132103398453e-11; /* 0xBDC5E43D, 0x693FB3C8 */
const V01: f64 = 1.27304834834123699328e-02; /* 0x3F8A1270, 0x91C9C71A */
const V02: f64 = 7.60068627350353253702e-05; /* 0x3F13ECBB, 0xF578C6C1 */
const V03: f64 = 2.59150851840457805467e-07; /* 0x3E91642D, 0x7FF202FD */
const V04: f64 = 4.41110311332675467403e-10; /* 0x3DFE5018, 0x3BD6D9EF */
pub fn y0(x: f64) -> f64 {
let z: f64;
let u: f64;
let v: f64;
let ix: u32;
let lx: u32;
ix = get_high_word(x);
lx = get_low_word(x);
/* y0(nan)=nan, y0(<0)=nan, y0(0)=-inf, y0(inf)=0 */
if ((ix << 1) | lx) == 0 {
return -1.0 / 0.0;
}
if (ix >> 31) != 0 {
return 0.0 / 0.0;
}
if ix >= 0x7ff00000 {
return 1.0 / x;
}
if ix >= 0x40000000 {
/* x >= 2 */
/* large ulp errors near zeros: 3.958, 7.086,.. */
return common(ix, x, true);
}
/* U(x^2)/V(x^2) + (2/pi)*j0(x)*log(x) */
if ix >= 0x3e400000 {
/* x >= 2**-27 */
/* large ulp error near the first zero, x ~= 0.89 */
z = x * x;
u = U00 + z * (U01 + z * (U02 + z * (U03 + z * (U04 + z * (U05 + z * U06)))));
v = 1.0 + z * (V01 + z * (V02 + z * (V03 + z * V04)));
return u / v + TPI * (j0(x) * log(x));
}
return U00 + TPI * log(x);
}
/* The asymptotic expansions of pzero is
* 1 - 9/128 s^2 + 11025/98304 s^4 - ..., where s = 1/x.
* For x >= 2, We approximate pzero by
* pzero(x) = 1 + (R/S)
* where R = pR0 + pR1*s^2 + pR2*s^4 + ... + pR5*s^10
* S = 1 + pS0*s^2 + ... + pS4*s^10
* and
* | pzero(x)-1-R/S | <= 2 ** ( -60.26)
*/
const PR8: [f64; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.00000000000000000000e+00, /* 0x00000000, 0x00000000 */
-7.03124999999900357484e-02, /* 0xBFB1FFFF, 0xFFFFFD32 */
-8.08167041275349795626e+00, /* 0xC02029D0, 0xB44FA779 */
-2.57063105679704847262e+02, /* 0xC0701102, 0x7B19E863 */
-2.48521641009428822144e+03, /* 0xC0A36A6E, 0xCD4DCAFC */
-5.25304380490729545272e+03, /* 0xC0B4850B, 0x36CC643D */
];
const PS8: [f64; 5] = [
1.16534364619668181717e+02, /* 0x405D2233, 0x07A96751 */
3.83374475364121826715e+03, /* 0x40ADF37D, 0x50596938 */
4.05978572648472545552e+04, /* 0x40E3D2BB, 0x6EB6B05F */
1.16752972564375915681e+05, /* 0x40FC810F, 0x8F9FA9BD */
4.76277284146730962675e+04, /* 0x40E74177, 0x4F2C49DC */
];
const PR5: [f64; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
-1.14125464691894502584e-11, /* 0xBDA918B1, 0x47E495CC */
-7.03124940873599280078e-02, /* 0xBFB1FFFF, 0xE69AFBC6 */
-4.15961064470587782438e+00, /* 0xC010A370, 0xF90C6BBF */
-6.76747652265167261021e+01, /* 0xC050EB2F, 0x5A7D1783 */
-3.31231299649172967747e+02, /* 0xC074B3B3, 0x6742CC63 */
-3.46433388365604912451e+02, /* 0xC075A6EF, 0x28A38BD7 */
];
const PS5: [f64; 5] = [
6.07539382692300335975e+01, /* 0x404E6081, 0x0C98C5DE */
1.05125230595704579173e+03, /* 0x40906D02, 0x5C7E2864 */
5.97897094333855784498e+03, /* 0x40B75AF8, 0x8FBE1D60 */
9.62544514357774460223e+03, /* 0x40C2CCB8, 0xFA76FA38 */
2.40605815922939109441e+03, /* 0x40A2CC1D, 0xC70BE864 */
];
const PR3: [f64; 6] = [
/* for x in [4.547,2.8571]=1/[0.2199,0.35001] */
-2.54704601771951915620e-09, /* 0xBE25E103, 0x6FE1AA86 */
-7.03119616381481654654e-02, /* 0xBFB1FFF6, 0xF7C0E24B */
-2.40903221549529611423e+00, /* 0xC00345B2, 0xAEA48074 */
-2.19659774734883086467e+01, /* 0xC035F74A, 0x4CB94E14 */
-5.80791704701737572236e+01, /* 0xC04D0A22, 0x420A1A45 */
-3.14479470594888503854e+01, /* 0xC03F72AC, 0xA892D80F */
];
const PS3: [f64; 5] = [
3.58560338055209726349e+01, /* 0x4041ED92, 0x84077DD3 */
3.61513983050303863820e+02, /* 0x40769839, 0x464A7C0E */
1.19360783792111533330e+03, /* 0x4092A66E, 0x6D1061D6 */
1.12799679856907414432e+03, /* 0x40919FFC, 0xB8C39B7E */
1.73580930813335754692e+02, /* 0x4065B296, 0xFC379081 */
];
const PR2: [f64; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
-8.87534333032526411254e-08, /* 0xBE77D316, 0xE927026D */
-7.03030995483624743247e-02, /* 0xBFB1FF62, 0x495E1E42 */
-1.45073846780952986357e+00, /* 0xBFF73639, 0x8A24A843 */
-7.63569613823527770791e+00, /* 0xC01E8AF3, 0xEDAFA7F3 */
-1.11931668860356747786e+01, /* 0xC02662E6, 0xC5246303 */
-3.23364579351335335033e+00, /* 0xC009DE81, 0xAF8FE70F */
];
const PS2: [f64; 5] = [
2.22202997532088808441e+01, /* 0x40363865, 0x908B5959 */
1.36206794218215208048e+02, /* 0x4061069E, 0x0EE8878F */
2.70470278658083486789e+02, /* 0x4070E786, 0x42EA079B */
1.53875394208320329881e+02, /* 0x40633C03, 0x3AB6FAFF */
1.46576176948256193810e+01, /* 0x402D50B3, 0x44391809 */
];
fn pzero(x: f64) -> f64 {
let p: &[f64; 6];
let q: &[f64; 5];
let z: f64;
let r: f64;
let s: f64;
let mut ix: u32;
ix = get_high_word(x);
ix &= 0x7fffffff;
if ix >= 0x40200000 {
p = &PR8;
q = &PS8;
} else if ix >= 0x40122E8B {
p = &PR5;
q = &PS5;
} else if ix >= 0x4006DB6D {
p = &PR3;
q = &PS3;
} else
/*ix >= 0x40000000*/
{
p = &PR2;
q = &PS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * q[4]))));
return 1.0 + r / s;
}
/* For x >= 8, the asymptotic expansions of qzero is
* -1/8 s + 75/1024 s^3 - ..., where s = 1/x.
* We approximate pzero by
* qzero(x) = s*(-1.25 + (R/S))
* where R = qR0 + qR1*s^2 + qR2*s^4 + ... + qR5*s^10
* S = 1 + qS0*s^2 + ... + qS5*s^12
* and
* | qzero(x)/s +1.25-R/S | <= 2 ** ( -61.22)
*/
const QR8: [f64; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.00000000000000000000e+00, /* 0x00000000, 0x00000000 */
7.32421874999935051953e-02, /* 0x3FB2BFFF, 0xFFFFFE2C */
1.17682064682252693899e+01, /* 0x40278952, 0x5BB334D6 */
5.57673380256401856059e+02, /* 0x40816D63, 0x15301825 */
8.85919720756468632317e+03, /* 0x40C14D99, 0x3E18F46D */
3.70146267776887834771e+04, /* 0x40E212D4, 0x0E901566 */
];
const QS8: [f64; 6] = [
1.63776026895689824414e+02, /* 0x406478D5, 0x365B39BC */
8.09834494656449805916e+03, /* 0x40BFA258, 0x4E6B0563 */
1.42538291419120476348e+05, /* 0x41016652, 0x54D38C3F */
8.03309257119514397345e+05, /* 0x412883DA, 0x83A52B43 */
8.40501579819060512818e+05, /* 0x4129A66B, 0x28DE0B3D */
-3.43899293537866615225e+05, /* 0xC114FD6D, 0x2C9530C5 */
];
const QR5: [f64; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
1.84085963594515531381e-11, /* 0x3DB43D8F, 0x29CC8CD9 */
7.32421766612684765896e-02, /* 0x3FB2BFFF, 0xD172B04C */
5.83563508962056953777e+00, /* 0x401757B0, 0xB9953DD3 */
1.35111577286449829671e+02, /* 0x4060E392, 0x0A8788E9 */
1.02724376596164097464e+03, /* 0x40900CF9, 0x9DC8C481 */
1.98997785864605384631e+03, /* 0x409F17E9, 0x53C6E3A6 */
];
const QS5: [f64; 6] = [
8.27766102236537761883e+01, /* 0x4054B1B3, 0xFB5E1543 */
2.07781416421392987104e+03, /* 0x40A03BA0, 0xDA21C0CE */
1.88472887785718085070e+04, /* 0x40D267D2, 0x7B591E6D */
5.67511122894947329769e+04, /* 0x40EBB5E3, 0x97E02372 */
3.59767538425114471465e+04, /* 0x40E19118, 0x1F7A54A0 */
-5.35434275601944773371e+03, /* 0xC0B4EA57, 0xBEDBC609 */
];
const QR3: [f64; 6] = [
/* for x in [4.547,2.8571]=1/[0.2199,0.35001] */
4.37741014089738620906e-09, /* 0x3E32CD03, 0x6ADECB82 */
7.32411180042911447163e-02, /* 0x3FB2BFEE, 0x0E8D0842 */
3.34423137516170720929e+00, /* 0x400AC0FC, 0x61149CF5 */
4.26218440745412650017e+01, /* 0x40454F98, 0x962DAEDD */
1.70808091340565596283e+02, /* 0x406559DB, 0xE25EFD1F */
1.66733948696651168575e+02, /* 0x4064D77C, 0x81FA21E0 */
];
const QS3: [f64; 6] = [
4.87588729724587182091e+01, /* 0x40486122, 0xBFE343A6 */
7.09689221056606015736e+02, /* 0x40862D83, 0x86544EB3 */
3.70414822620111362994e+03, /* 0x40ACF04B, 0xE44DFC63 */
6.46042516752568917582e+03, /* 0x40B93C6C, 0xD7C76A28 */
2.51633368920368957333e+03, /* 0x40A3A8AA, 0xD94FB1C0 */
-1.49247451836156386662e+02, /* 0xC062A7EB, 0x201CF40F */
];
const QR2: [f64; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
1.50444444886983272379e-07, /* 0x3E84313B, 0x54F76BDB */
7.32234265963079278272e-02, /* 0x3FB2BEC5, 0x3E883E34 */
1.99819174093815998816e+00, /* 0x3FFFF897, 0xE727779C */
1.44956029347885735348e+01, /* 0x402CFDBF, 0xAAF96FE5 */
3.16662317504781540833e+01, /* 0x403FAA8E, 0x29FBDC4A */
1.62527075710929267416e+01, /* 0x403040B1, 0x71814BB4 */
];
const QS2: [f64; 6] = [
3.03655848355219184498e+01, /* 0x403E5D96, 0xF7C07AED */
2.69348118608049844624e+02, /* 0x4070D591, 0xE4D14B40 */
8.44783757595320139444e+02, /* 0x408A6645, 0x22B3BF22 */
8.82935845112488550512e+02, /* 0x408B977C, 0x9C5CC214 */
2.12666388511798828631e+02, /* 0x406A9553, 0x0E001365 */
-5.31095493882666946917e+00, /* 0xC0153E6A, 0xF8B32931 */
];
fn qzero(x: f64) -> f64 {
let p: &[f64; 6];
let q: &[f64; 6];
let s: f64;
let r: f64;
let z: f64;
let mut ix: u32;
ix = get_high_word(x);
ix &= 0x7fffffff;
if ix >= 0x40200000 {
p = &QR8;
q = &QS8;
} else if ix >= 0x40122E8B {
p = &QR5;
q = &QS5;
} else if ix >= 0x4006DB6D {
p = &QR3;
q = &QS3;
} else
/*ix >= 0x40000000*/
{
p = &QR2;
q = &QS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * (q[4] + z * q[5])))));
return (-0.125 + r / s) / x;
}

View File

@ -0,0 +1,359 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_j0f.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{cosf, fabsf, logf, sinf, sqrtf};
const INVSQRTPI: f32 = 5.6418961287e-01; /* 0x3f106ebb */
const TPI: f32 = 6.3661974669e-01; /* 0x3f22f983 */
fn common(ix: u32, x: f32, y0: bool) -> f32 {
let z: f32;
let s: f32;
let mut c: f32;
let mut ss: f32;
let mut cc: f32;
/*
* j0(x) = 1/sqrt(pi) * (P(0,x)*cc - Q(0,x)*ss) / sqrt(x)
* y0(x) = 1/sqrt(pi) * (P(0,x)*ss + Q(0,x)*cc) / sqrt(x)
*/
s = sinf(x);
c = cosf(x);
if y0 {
c = -c;
}
cc = s + c;
if ix < 0x7f000000 {
ss = s - c;
z = -cosf(2.0 * x);
if s * c < 0.0 {
cc = z / ss;
} else {
ss = z / cc;
}
if ix < 0x58800000 {
if y0 {
ss = -ss;
}
cc = pzerof(x) * cc - qzerof(x) * ss;
}
}
return INVSQRTPI * cc / sqrtf(x);
}
/* R0/S0 on [0, 2.00] */
const R02: f32 = 1.5625000000e-02; /* 0x3c800000 */
const R03: f32 = -1.8997929874e-04; /* 0xb947352e */
const R04: f32 = 1.8295404516e-06; /* 0x35f58e88 */
const R05: f32 = -4.6183270541e-09; /* 0xb19eaf3c */
const S01: f32 = 1.5619102865e-02; /* 0x3c7fe744 */
const S02: f32 = 1.1692678527e-04; /* 0x38f53697 */
const S03: f32 = 5.1354652442e-07; /* 0x3509daa6 */
const S04: f32 = 1.1661400734e-09; /* 0x30a045e8 */
pub fn j0f(mut x: f32) -> f32 {
let z: f32;
let r: f32;
let s: f32;
let mut ix: u32;
ix = x.to_bits();
ix &= 0x7fffffff;
if ix >= 0x7f800000 {
return 1.0 / (x * x);
}
x = fabsf(x);
if ix >= 0x40000000 {
/* |x| >= 2 */
/* large ulp error near zeros */
return common(ix, x, false);
}
if ix >= 0x3a000000 {
/* |x| >= 2**-11 */
/* up to 4ulp error near 2 */
z = x * x;
r = z * (R02 + z * (R03 + z * (R04 + z * R05)));
s = 1.0 + z * (S01 + z * (S02 + z * (S03 + z * S04)));
return (1.0 + x / 2.0) * (1.0 - x / 2.0) + z * (r / s);
}
if ix >= 0x21800000 {
/* |x| >= 2**-60 */
x = 0.25 * x * x;
}
return 1.0 - x;
}
const U00: f32 = -7.3804296553e-02; /* 0xbd9726b5 */
const U01: f32 = 1.7666645348e-01; /* 0x3e34e80d */
const U02: f32 = -1.3818567619e-02; /* 0xbc626746 */
const U03: f32 = 3.4745343146e-04; /* 0x39b62a69 */
const U04: f32 = -3.8140706238e-06; /* 0xb67ff53c */
const U05: f32 = 1.9559013964e-08; /* 0x32a802ba */
const U06: f32 = -3.9820518410e-11; /* 0xae2f21eb */
const V01: f32 = 1.2730483897e-02; /* 0x3c509385 */
const V02: f32 = 7.6006865129e-05; /* 0x389f65e0 */
const V03: f32 = 2.5915085189e-07; /* 0x348b216c */
const V04: f32 = 4.4111031494e-10; /* 0x2ff280c2 */
pub fn y0f(x: f32) -> f32 {
let z: f32;
let u: f32;
let v: f32;
let ix: u32;
ix = x.to_bits();
if (ix & 0x7fffffff) == 0 {
return -1.0 / 0.0;
}
if (ix >> 31) != 0 {
return 0.0 / 0.0;
}
if ix >= 0x7f800000 {
return 1.0 / x;
}
if ix >= 0x40000000 {
/* |x| >= 2.0 */
/* large ulp error near zeros */
return common(ix, x, true);
}
if ix >= 0x39000000 {
/* x >= 2**-13 */
/* large ulp error at x ~= 0.89 */
z = x * x;
u = U00 + z * (U01 + z * (U02 + z * (U03 + z * (U04 + z * (U05 + z * U06)))));
v = 1.0 + z * (V01 + z * (V02 + z * (V03 + z * V04)));
return u / v + TPI * (j0f(x) * logf(x));
}
return U00 + TPI * logf(x);
}
/* The asymptotic expansions of pzero is
* 1 - 9/128 s^2 + 11025/98304 s^4 - ..., where s = 1/x.
* For x >= 2, We approximate pzero by
* pzero(x) = 1 + (R/S)
* where R = pR0 + pR1*s^2 + pR2*s^4 + ... + pR5*s^10
* S = 1 + pS0*s^2 + ... + pS4*s^10
* and
* | pzero(x)-1-R/S | <= 2 ** ( -60.26)
*/
const PR8: [f32; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.0000000000e+00, /* 0x00000000 */
-7.0312500000e-02, /* 0xbd900000 */
-8.0816707611e+00, /* 0xc1014e86 */
-2.5706311035e+02, /* 0xc3808814 */
-2.4852163086e+03, /* 0xc51b5376 */
-5.2530439453e+03, /* 0xc5a4285a */
];
const PS8: [f32; 5] = [
1.1653436279e+02, /* 0x42e91198 */
3.8337448730e+03, /* 0x456f9beb */
4.0597855469e+04, /* 0x471e95db */
1.1675296875e+05, /* 0x47e4087c */
4.7627726562e+04, /* 0x473a0bba */
];
const PR5: [f32; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
-1.1412546255e-11, /* 0xad48c58a */
-7.0312492549e-02, /* 0xbd8fffff */
-4.1596107483e+00, /* 0xc0851b88 */
-6.7674766541e+01, /* 0xc287597b */
-3.3123129272e+02, /* 0xc3a59d9b */
-3.4643338013e+02, /* 0xc3ad3779 */
];
const PS5: [f32; 5] = [
6.0753936768e+01, /* 0x42730408 */
1.0512523193e+03, /* 0x44836813 */
5.9789707031e+03, /* 0x45bad7c4 */
9.6254453125e+03, /* 0x461665c8 */
2.4060581055e+03, /* 0x451660ee */
];
const PR3: [f32; 6] = [
/* for x in [4.547,2.8571]=1/[0.2199,0.35001] */
-2.5470459075e-09, /* 0xb12f081b */
-7.0311963558e-02, /* 0xbd8fffb8 */
-2.4090321064e+00, /* 0xc01a2d95 */
-2.1965976715e+01, /* 0xc1afba52 */
-5.8079170227e+01, /* 0xc2685112 */
-3.1447946548e+01, /* 0xc1fb9565 */
];
const PS3: [f32; 5] = [
3.5856033325e+01, /* 0x420f6c94 */
3.6151397705e+02, /* 0x43b4c1ca */
1.1936077881e+03, /* 0x44953373 */
1.1279968262e+03, /* 0x448cffe6 */
1.7358093262e+02, /* 0x432d94b8 */
];
const PR2: [f32; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
-8.8753431271e-08, /* 0xb3be98b7 */
-7.0303097367e-02, /* 0xbd8ffb12 */
-1.4507384300e+00, /* 0xbfb9b1cc */
-7.6356959343e+00, /* 0xc0f4579f */
-1.1193166733e+01, /* 0xc1331736 */
-3.2336456776e+00, /* 0xc04ef40d */
];
const PS2: [f32; 5] = [
2.2220300674e+01, /* 0x41b1c32d */
1.3620678711e+02, /* 0x430834f0 */
2.7047027588e+02, /* 0x43873c32 */
1.5387539673e+02, /* 0x4319e01a */
1.4657617569e+01, /* 0x416a859a */
];
fn pzerof(x: f32) -> f32 {
let p: &[f32; 6];
let q: &[f32; 5];
let z: f32;
let r: f32;
let s: f32;
let mut ix: u32;
ix = x.to_bits();
ix &= 0x7fffffff;
if ix >= 0x41000000 {
p = &PR8;
q = &PS8;
} else if ix >= 0x409173eb {
p = &PR5;
q = &PS5;
} else if ix >= 0x4036d917 {
p = &PR3;
q = &PS3;
} else
/*ix >= 0x40000000*/
{
p = &PR2;
q = &PS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * q[4]))));
return 1.0 + r / s;
}
/* For x >= 8, the asymptotic expansions of qzero is
* -1/8 s + 75/1024 s^3 - ..., where s = 1/x.
* We approximate pzero by
* qzero(x) = s*(-1.25 + (R/S))
* where R = qR0 + qR1*s^2 + qR2*s^4 + ... + qR5*s^10
* S = 1 + qS0*s^2 + ... + qS5*s^12
* and
* | qzero(x)/s +1.25-R/S | <= 2 ** ( -61.22)
*/
const QR8: [f32; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.0000000000e+00, /* 0x00000000 */
7.3242187500e-02, /* 0x3d960000 */
1.1768206596e+01, /* 0x413c4a93 */
5.5767340088e+02, /* 0x440b6b19 */
8.8591972656e+03, /* 0x460a6cca */
3.7014625000e+04, /* 0x471096a0 */
];
const QS8: [f32; 6] = [
1.6377603149e+02, /* 0x4323c6aa */
8.0983447266e+03, /* 0x45fd12c2 */
1.4253829688e+05, /* 0x480b3293 */
8.0330925000e+05, /* 0x49441ed4 */
8.4050156250e+05, /* 0x494d3359 */
-3.4389928125e+05, /* 0xc8a7eb69 */
];
const QR5: [f32; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
1.8408595828e-11, /* 0x2da1ec79 */
7.3242180049e-02, /* 0x3d95ffff */
5.8356351852e+00, /* 0x40babd86 */
1.3511157227e+02, /* 0x43071c90 */
1.0272437744e+03, /* 0x448067cd */
1.9899779053e+03, /* 0x44f8bf4b */
];
const QS5: [f32; 6] = [
8.2776611328e+01, /* 0x42a58da0 */
2.0778142090e+03, /* 0x4501dd07 */
1.8847289062e+04, /* 0x46933e94 */
5.6751113281e+04, /* 0x475daf1d */
3.5976753906e+04, /* 0x470c88c1 */
-5.3543427734e+03, /* 0xc5a752be */
];
const QR3: [f32; 6] = [
/* for x in [4.547,2.8571]=1/[0.2199,0.35001] */
4.3774099900e-09, /* 0x3196681b */
7.3241114616e-02, /* 0x3d95ff70 */
3.3442313671e+00, /* 0x405607e3 */
4.2621845245e+01, /* 0x422a7cc5 */
1.7080809021e+02, /* 0x432acedf */
1.6673394775e+02, /* 0x4326bbe4 */
];
const QS3: [f32; 6] = [
4.8758872986e+01, /* 0x42430916 */
7.0968920898e+02, /* 0x44316c1c */
3.7041481934e+03, /* 0x4567825f */
6.4604252930e+03, /* 0x45c9e367 */
2.5163337402e+03, /* 0x451d4557 */
-1.4924745178e+02, /* 0xc3153f59 */
];
const QR2: [f32; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
1.5044444979e-07, /* 0x342189db */
7.3223426938e-02, /* 0x3d95f62a */
1.9981917143e+00, /* 0x3fffc4bf */
1.4495602608e+01, /* 0x4167edfd */
3.1666231155e+01, /* 0x41fd5471 */
1.6252708435e+01, /* 0x4182058c */
];
const QS2: [f32; 6] = [
3.0365585327e+01, /* 0x41f2ecb8 */
2.6934811401e+02, /* 0x4386ac8f */
8.4478375244e+02, /* 0x44533229 */
8.8293585205e+02, /* 0x445cbbe5 */
2.1266638184e+02, /* 0x4354aa98 */
-5.3109550476e+00, /* 0xc0a9f358 */
];
fn qzerof(x: f32) -> f32 {
let p: &[f32; 6];
let q: &[f32; 6];
let s: f32;
let r: f32;
let z: f32;
let mut ix: u32;
ix = x.to_bits();
ix &= 0x7fffffff;
if ix >= 0x41000000 {
p = &QR8;
q = &QS8;
} else if ix >= 0x409173eb {
p = &QR5;
q = &QS5;
} else if ix >= 0x4036d917 {
p = &QR3;
q = &QS3;
} else
/*ix >= 0x40000000*/
{
p = &QR2;
q = &QS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * (q[4] + z * q[5])))));
return (-0.125 + r / s) / x;
}

View File

@ -0,0 +1,414 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_j1.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* j1(x), y1(x)
* Bessel function of the first and second kinds of order zero.
* Method -- j1(x):
* 1. For tiny x, we use j1(x) = x/2 - x^3/16 + x^5/384 - ...
* 2. Reduce x to |x| since j1(x)=-j1(-x), and
* for x in (0,2)
* j1(x) = x/2 + x*z*R0/S0, where z = x*x;
* (precision: |j1/x - 1/2 - R0/S0 |<2**-61.51 )
* for x in (2,inf)
* j1(x) = sqrt(2/(pi*x))*(p1(x)*cos(x1)-q1(x)*sin(x1))
* y1(x) = sqrt(2/(pi*x))*(p1(x)*sin(x1)+q1(x)*cos(x1))
* where x1 = x-3*pi/4. It is better to compute sin(x1),cos(x1)
* as follow:
* cos(x1) = cos(x)cos(3pi/4)+sin(x)sin(3pi/4)
* = 1/sqrt(2) * (sin(x) - cos(x))
* sin(x1) = sin(x)cos(3pi/4)-cos(x)sin(3pi/4)
* = -1/sqrt(2) * (sin(x) + cos(x))
* (To avoid cancellation, use
* sin(x) +- cos(x) = -cos(2x)/(sin(x) -+ cos(x))
* to compute the worse one.)
*
* 3 Special cases
* j1(nan)= nan
* j1(0) = 0
* j1(inf) = 0
*
* Method -- y1(x):
* 1. screen out x<=0 cases: y1(0)=-inf, y1(x<0)=NaN
* 2. For x<2.
* Since
* y1(x) = 2/pi*(j1(x)*(ln(x/2)+Euler)-1/x-x/2+5/64*x^3-...)
* therefore y1(x)-2/pi*j1(x)*ln(x)-1/x is an odd function.
* We use the following function to approximate y1,
* y1(x) = x*U(z)/V(z) + (2/pi)*(j1(x)*ln(x)-1/x), z= x^2
* where for x in [0,2] (abs err less than 2**-65.89)
* U(z) = U0[0] + U0[1]*z + ... + U0[4]*z^4
* V(z) = 1 + v0[0]*z + ... + v0[4]*z^5
* Note: For tiny x, 1/x dominate y1 and hence
* y1(tiny) = -2/pi/tiny, (choose tiny<2**-54)
* 3. For x>=2.
* y1(x) = sqrt(2/(pi*x))*(p1(x)*sin(x1)+q1(x)*cos(x1))
* where x1 = x-3*pi/4. It is better to compute sin(x1),cos(x1)
* by method mentioned above.
*/
use super::{cos, fabs, get_high_word, get_low_word, log, sin, sqrt};
const INVSQRTPI: f64 = 5.64189583547756279280e-01; /* 0x3FE20DD7, 0x50429B6D */
const TPI: f64 = 6.36619772367581382433e-01; /* 0x3FE45F30, 0x6DC9C883 */
fn common(ix: u32, x: f64, y1: bool, sign: bool) -> f64 {
let z: f64;
let mut s: f64;
let c: f64;
let mut ss: f64;
let mut cc: f64;
/*
* j1(x) = sqrt(2/(pi*x))*(p1(x)*cos(x-3pi/4)-q1(x)*sin(x-3pi/4))
* y1(x) = sqrt(2/(pi*x))*(p1(x)*sin(x-3pi/4)+q1(x)*cos(x-3pi/4))
*
* sin(x-3pi/4) = -(sin(x) + cos(x))/sqrt(2)
* cos(x-3pi/4) = (sin(x) - cos(x))/sqrt(2)
* sin(x) +- cos(x) = -cos(2x)/(sin(x) -+ cos(x))
*/
s = sin(x);
if y1 {
s = -s;
}
c = cos(x);
cc = s - c;
if ix < 0x7fe00000 {
/* avoid overflow in 2*x */
ss = -s - c;
z = cos(2.0 * x);
if s * c > 0.0 {
cc = z / ss;
} else {
ss = z / cc;
}
if ix < 0x48000000 {
if y1 {
ss = -ss;
}
cc = pone(x) * cc - qone(x) * ss;
}
}
if sign {
cc = -cc;
}
return INVSQRTPI * cc / sqrt(x);
}
/* R0/S0 on [0,2] */
const R00: f64 = -6.25000000000000000000e-02; /* 0xBFB00000, 0x00000000 */
const R01: f64 = 1.40705666955189706048e-03; /* 0x3F570D9F, 0x98472C61 */
const R02: f64 = -1.59955631084035597520e-05; /* 0xBEF0C5C6, 0xBA169668 */
const R03: f64 = 4.96727999609584448412e-08; /* 0x3E6AAAFA, 0x46CA0BD9 */
const S01: f64 = 1.91537599538363460805e-02; /* 0x3F939D0B, 0x12637E53 */
const S02: f64 = 1.85946785588630915560e-04; /* 0x3F285F56, 0xB9CDF664 */
const S03: f64 = 1.17718464042623683263e-06; /* 0x3EB3BFF8, 0x333F8498 */
const S04: f64 = 5.04636257076217042715e-09; /* 0x3E35AC88, 0xC97DFF2C */
const S05: f64 = 1.23542274426137913908e-11; /* 0x3DAB2ACF, 0xCFB97ED8 */
pub fn j1(x: f64) -> f64 {
let mut z: f64;
let r: f64;
let s: f64;
let mut ix: u32;
let sign: bool;
ix = get_high_word(x);
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix >= 0x7ff00000 {
return 1.0 / (x * x);
}
if ix >= 0x40000000 {
/* |x| >= 2 */
return common(ix, fabs(x), false, sign);
}
if ix >= 0x38000000 {
/* |x| >= 2**-127 */
z = x * x;
r = z * (R00 + z * (R01 + z * (R02 + z * R03)));
s = 1.0 + z * (S01 + z * (S02 + z * (S03 + z * (S04 + z * S05))));
z = r / s;
} else {
/* avoid underflow, raise inexact if x!=0 */
z = x;
}
return (0.5 + z) * x;
}
const U0: [f64; 5] = [
-1.96057090646238940668e-01, /* 0xBFC91866, 0x143CBC8A */
5.04438716639811282616e-02, /* 0x3FA9D3C7, 0x76292CD1 */
-1.91256895875763547298e-03, /* 0xBF5F55E5, 0x4844F50F */
2.35252600561610495928e-05, /* 0x3EF8AB03, 0x8FA6B88E */
-9.19099158039878874504e-08, /* 0xBE78AC00, 0x569105B8 */
];
const V0: [f64; 5] = [
1.99167318236649903973e-02, /* 0x3F94650D, 0x3F4DA9F0 */
2.02552581025135171496e-04, /* 0x3F2A8C89, 0x6C257764 */
1.35608801097516229404e-06, /* 0x3EB6C05A, 0x894E8CA6 */
6.22741452364621501295e-09, /* 0x3E3ABF1D, 0x5BA69A86 */
1.66559246207992079114e-11, /* 0x3DB25039, 0xDACA772A */
];
pub fn y1(x: f64) -> f64 {
let z: f64;
let u: f64;
let v: f64;
let ix: u32;
let lx: u32;
ix = get_high_word(x);
lx = get_low_word(x);
/* y1(nan)=nan, y1(<0)=nan, y1(0)=-inf, y1(inf)=0 */
if (ix << 1 | lx) == 0 {
return -1.0 / 0.0;
}
if (ix >> 31) != 0 {
return 0.0 / 0.0;
}
if ix >= 0x7ff00000 {
return 1.0 / x;
}
if ix >= 0x40000000 {
/* x >= 2 */
return common(ix, x, true, false);
}
if ix < 0x3c900000 {
/* x < 2**-54 */
return -TPI / x;
}
z = x * x;
u = U0[0] + z * (U0[1] + z * (U0[2] + z * (U0[3] + z * U0[4])));
v = 1.0 + z * (V0[0] + z * (V0[1] + z * (V0[2] + z * (V0[3] + z * V0[4]))));
return x * (u / v) + TPI * (j1(x) * log(x) - 1.0 / x);
}
/* For x >= 8, the asymptotic expansions of pone is
* 1 + 15/128 s^2 - 4725/2^15 s^4 - ..., where s = 1/x.
* We approximate pone by
* pone(x) = 1 + (R/S)
* where R = pr0 + pr1*s^2 + pr2*s^4 + ... + pr5*s^10
* S = 1 + ps0*s^2 + ... + ps4*s^10
* and
* | pone(x)-1-R/S | <= 2 ** ( -60.06)
*/
const PR8: [f64; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.00000000000000000000e+00, /* 0x00000000, 0x00000000 */
1.17187499999988647970e-01, /* 0x3FBDFFFF, 0xFFFFFCCE */
1.32394806593073575129e+01, /* 0x402A7A9D, 0x357F7FCE */
4.12051854307378562225e+02, /* 0x4079C0D4, 0x652EA590 */
3.87474538913960532227e+03, /* 0x40AE457D, 0xA3A532CC */
7.91447954031891731574e+03, /* 0x40BEEA7A, 0xC32782DD */
];
const PS8: [f64; 5] = [
1.14207370375678408436e+02, /* 0x405C8D45, 0x8E656CAC */
3.65093083420853463394e+03, /* 0x40AC85DC, 0x964D274F */
3.69562060269033463555e+04, /* 0x40E20B86, 0x97C5BB7F */
9.76027935934950801311e+04, /* 0x40F7D42C, 0xB28F17BB */
3.08042720627888811578e+04, /* 0x40DE1511, 0x697A0B2D */
];
const PR5: [f64; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
1.31990519556243522749e-11, /* 0x3DAD0667, 0xDAE1CA7D */
1.17187493190614097638e-01, /* 0x3FBDFFFF, 0xE2C10043 */
6.80275127868432871736e+00, /* 0x401B3604, 0x6E6315E3 */
1.08308182990189109773e+02, /* 0x405B13B9, 0x452602ED */
5.17636139533199752805e+02, /* 0x40802D16, 0xD052D649 */
5.28715201363337541807e+02, /* 0x408085B8, 0xBB7E0CB7 */
];
const PS5: [f64; 5] = [
5.92805987221131331921e+01, /* 0x404DA3EA, 0xA8AF633D */
9.91401418733614377743e+02, /* 0x408EFB36, 0x1B066701 */
5.35326695291487976647e+03, /* 0x40B4E944, 0x5706B6FB */
7.84469031749551231769e+03, /* 0x40BEA4B0, 0xB8A5BB15 */
1.50404688810361062679e+03, /* 0x40978030, 0x036F5E51 */
];
const PR3: [f64; 6] = [
3.02503916137373618024e-09, /* 0x3E29FC21, 0xA7AD9EDD */
1.17186865567253592491e-01, /* 0x3FBDFFF5, 0x5B21D17B */
3.93297750033315640650e+00, /* 0x400F76BC, 0xE85EAD8A */
3.51194035591636932736e+01, /* 0x40418F48, 0x9DA6D129 */
9.10550110750781271918e+01, /* 0x4056C385, 0x4D2C1837 */
4.85590685197364919645e+01, /* 0x4048478F, 0x8EA83EE5 */
];
const PS3: [f64; 5] = [
3.47913095001251519989e+01, /* 0x40416549, 0xA134069C */
3.36762458747825746741e+02, /* 0x40750C33, 0x07F1A75F */
1.04687139975775130551e+03, /* 0x40905B7C, 0x5037D523 */
8.90811346398256432622e+02, /* 0x408BD67D, 0xA32E31E9 */
1.03787932439639277504e+02, /* 0x4059F26D, 0x7C2EED53 */
];
const PR2: [f64; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
1.07710830106873743082e-07, /* 0x3E7CE9D4, 0xF65544F4 */
1.17176219462683348094e-01, /* 0x3FBDFF42, 0xBE760D83 */
2.36851496667608785174e+00, /* 0x4002F2B7, 0xF98FAEC0 */
1.22426109148261232917e+01, /* 0x40287C37, 0x7F71A964 */
1.76939711271687727390e+01, /* 0x4031B1A8, 0x177F8EE2 */
5.07352312588818499250e+00, /* 0x40144B49, 0xA574C1FE */
];
const PS2: [f64; 5] = [
2.14364859363821409488e+01, /* 0x40356FBD, 0x8AD5ECDC */
1.25290227168402751090e+02, /* 0x405F5293, 0x14F92CD5 */
2.32276469057162813669e+02, /* 0x406D08D8, 0xD5A2DBD9 */
1.17679373287147100768e+02, /* 0x405D6B7A, 0xDA1884A9 */
8.36463893371618283368e+00, /* 0x4020BAB1, 0xF44E5192 */
];
fn pone(x: f64) -> f64 {
let p: &[f64; 6];
let q: &[f64; 5];
let z: f64;
let r: f64;
let s: f64;
let mut ix: u32;
ix = get_high_word(x);
ix &= 0x7fffffff;
if ix >= 0x40200000 {
p = &PR8;
q = &PS8;
} else if ix >= 0x40122E8B {
p = &PR5;
q = &PS5;
} else if ix >= 0x4006DB6D {
p = &PR3;
q = &PS3;
} else
/*ix >= 0x40000000*/
{
p = &PR2;
q = &PS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * q[4]))));
return 1.0 + r / s;
}
/* For x >= 8, the asymptotic expansions of qone is
* 3/8 s - 105/1024 s^3 - ..., where s = 1/x.
* We approximate pone by
* qone(x) = s*(0.375 + (R/S))
* where R = qr1*s^2 + qr2*s^4 + ... + qr5*s^10
* S = 1 + qs1*s^2 + ... + qs6*s^12
* and
* | qone(x)/s -0.375-R/S | <= 2 ** ( -61.13)
*/
const QR8: [f64; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.00000000000000000000e+00, /* 0x00000000, 0x00000000 */
-1.02539062499992714161e-01, /* 0xBFBA3FFF, 0xFFFFFDF3 */
-1.62717534544589987888e+01, /* 0xC0304591, 0xA26779F7 */
-7.59601722513950107896e+02, /* 0xC087BCD0, 0x53E4B576 */
-1.18498066702429587167e+04, /* 0xC0C724E7, 0x40F87415 */
-4.84385124285750353010e+04, /* 0xC0E7A6D0, 0x65D09C6A */
];
const QS8: [f64; 6] = [
1.61395369700722909556e+02, /* 0x40642CA6, 0xDE5BCDE5 */
7.82538599923348465381e+03, /* 0x40BE9162, 0xD0D88419 */
1.33875336287249578163e+05, /* 0x4100579A, 0xB0B75E98 */
7.19657723683240939863e+05, /* 0x4125F653, 0x72869C19 */
6.66601232617776375264e+05, /* 0x412457D2, 0x7719AD5C */
-2.94490264303834643215e+05, /* 0xC111F969, 0x0EA5AA18 */
];
const QR5: [f64; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
-2.08979931141764104297e-11, /* 0xBDB6FA43, 0x1AA1A098 */
-1.02539050241375426231e-01, /* 0xBFBA3FFF, 0xCB597FEF */
-8.05644828123936029840e+00, /* 0xC0201CE6, 0xCA03AD4B */
-1.83669607474888380239e+02, /* 0xC066F56D, 0x6CA7B9B0 */
-1.37319376065508163265e+03, /* 0xC09574C6, 0x6931734F */
-2.61244440453215656817e+03, /* 0xC0A468E3, 0x88FDA79D */
];
const QS5: [f64; 6] = [
8.12765501384335777857e+01, /* 0x405451B2, 0xFF5A11B2 */
1.99179873460485964642e+03, /* 0x409F1F31, 0xE77BF839 */
1.74684851924908907677e+04, /* 0x40D10F1F, 0x0D64CE29 */
4.98514270910352279316e+04, /* 0x40E8576D, 0xAABAD197 */
2.79480751638918118260e+04, /* 0x40DB4B04, 0xCF7C364B */
-4.71918354795128470869e+03, /* 0xC0B26F2E, 0xFCFFA004 */
];
const QR3: [f64; 6] = [
-5.07831226461766561369e-09, /* 0xBE35CFA9, 0xD38FC84F */
-1.02537829820837089745e-01, /* 0xBFBA3FEB, 0x51AEED54 */
-4.61011581139473403113e+00, /* 0xC01270C2, 0x3302D9FF */
-5.78472216562783643212e+01, /* 0xC04CEC71, 0xC25D16DA */
-2.28244540737631695038e+02, /* 0xC06C87D3, 0x4718D55F */
-2.19210128478909325622e+02, /* 0xC06B66B9, 0x5F5C1BF6 */
];
const QS3: [f64; 6] = [
4.76651550323729509273e+01, /* 0x4047D523, 0xCCD367E4 */
6.73865112676699709482e+02, /* 0x40850EEB, 0xC031EE3E */
3.38015286679526343505e+03, /* 0x40AA684E, 0x448E7C9A */
5.54772909720722782367e+03, /* 0x40B5ABBA, 0xA61D54A6 */
1.90311919338810798763e+03, /* 0x409DBC7A, 0x0DD4DF4B */
-1.35201191444307340817e+02, /* 0xC060E670, 0x290A311F */
];
const QR2: [f64; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
-1.78381727510958865572e-07, /* 0xBE87F126, 0x44C626D2 */
-1.02517042607985553460e-01, /* 0xBFBA3E8E, 0x9148B010 */
-2.75220568278187460720e+00, /* 0xC0060484, 0x69BB4EDA */
-1.96636162643703720221e+01, /* 0xC033A9E2, 0xC168907F */
-4.23253133372830490089e+01, /* 0xC04529A3, 0xDE104AAA */
-2.13719211703704061733e+01, /* 0xC0355F36, 0x39CF6E52 */
];
const QS2: [f64; 6] = [
2.95333629060523854548e+01, /* 0x403D888A, 0x78AE64FF */
2.52981549982190529136e+02, /* 0x406F9F68, 0xDB821CBA */
7.57502834868645436472e+02, /* 0x4087AC05, 0xCE49A0F7 */
7.39393205320467245656e+02, /* 0x40871B25, 0x48D4C029 */
1.55949003336666123687e+02, /* 0x40637E5E, 0x3C3ED8D4 */
-4.95949898822628210127e+00, /* 0xC013D686, 0xE71BE86B */
];
fn qone(x: f64) -> f64 {
let p: &[f64; 6];
let q: &[f64; 6];
let s: f64;
let r: f64;
let z: f64;
let mut ix: u32;
ix = get_high_word(x);
ix &= 0x7fffffff;
if ix >= 0x40200000 {
p = &QR8;
q = &QS8;
} else if ix >= 0x40122E8B {
p = &QR5;
q = &QS5;
} else if ix >= 0x4006DB6D {
p = &QR3;
q = &QS3;
} else
/*ix >= 0x40000000*/
{
p = &QR2;
q = &QS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * (q[4] + z * q[5])))));
return (0.375 + r / s) / x;
}

View File

@ -0,0 +1,380 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_j1f.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{cosf, fabsf, logf, sinf, sqrtf};
const INVSQRTPI: f32 = 5.6418961287e-01; /* 0x3f106ebb */
const TPI: f32 = 6.3661974669e-01; /* 0x3f22f983 */
fn common(ix: u32, x: f32, y1: bool, sign: bool) -> f32 {
let z: f64;
let mut s: f64;
let c: f64;
let mut ss: f64;
let mut cc: f64;
s = sinf(x) as f64;
if y1 {
s = -s;
}
c = cosf(x) as f64;
cc = s - c;
if ix < 0x7f000000 {
ss = -s - c;
z = cosf(2.0 * x) as f64;
if s * c > 0.0 {
cc = z / ss;
} else {
ss = z / cc;
}
if ix < 0x58800000 {
if y1 {
ss = -ss;
}
cc = (ponef(x) as f64) * cc - (qonef(x) as f64) * ss;
}
}
if sign {
cc = -cc;
}
return (((INVSQRTPI as f64) * cc) / (sqrtf(x) as f64)) as f32;
}
/* R0/S0 on [0,2] */
const R00: f32 = -6.2500000000e-02; /* 0xbd800000 */
const R01: f32 = 1.4070566976e-03; /* 0x3ab86cfd */
const R02: f32 = -1.5995563444e-05; /* 0xb7862e36 */
const R03: f32 = 4.9672799207e-08; /* 0x335557d2 */
const S01: f32 = 1.9153760746e-02; /* 0x3c9ce859 */
const S02: f32 = 1.8594678841e-04; /* 0x3942fab6 */
const S03: f32 = 1.1771846857e-06; /* 0x359dffc2 */
const S04: f32 = 5.0463624390e-09; /* 0x31ad6446 */
const S05: f32 = 1.2354227016e-11; /* 0x2d59567e */
pub fn j1f(x: f32) -> f32 {
let mut z: f32;
let r: f32;
let s: f32;
let mut ix: u32;
let sign: bool;
ix = x.to_bits();
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix >= 0x7f800000 {
return 1.0 / (x * x);
}
if ix >= 0x40000000 {
/* |x| >= 2 */
return common(ix, fabsf(x), false, sign);
}
if ix >= 0x39000000 {
/* |x| >= 2**-13 */
z = x * x;
r = z * (R00 + z * (R01 + z * (R02 + z * R03)));
s = 1.0 + z * (S01 + z * (S02 + z * (S03 + z * (S04 + z * S05))));
z = 0.5 + r / s;
} else {
z = 0.5;
}
return z * x;
}
const U0: [f32; 5] = [
-1.9605709612e-01, /* 0xbe48c331 */
5.0443872809e-02, /* 0x3d4e9e3c */
-1.9125689287e-03, /* 0xbafaaf2a */
2.3525259166e-05, /* 0x37c5581c */
-9.1909917899e-08, /* 0xb3c56003 */
];
const V0: [f32; 5] = [
1.9916731864e-02, /* 0x3ca3286a */
2.0255257550e-04, /* 0x3954644b */
1.3560879779e-06, /* 0x35b602d4 */
6.2274145840e-09, /* 0x31d5f8eb */
1.6655924903e-11, /* 0x2d9281cf */
];
pub fn y1f(x: f32) -> f32 {
let z: f32;
let u: f32;
let v: f32;
let ix: u32;
ix = x.to_bits();
if (ix & 0x7fffffff) == 0 {
return -1.0 / 0.0;
}
if (ix >> 31) != 0 {
return 0.0 / 0.0;
}
if ix >= 0x7f800000 {
return 1.0 / x;
}
if ix >= 0x40000000 {
/* |x| >= 2.0 */
return common(ix, x, true, false);
}
if ix < 0x33000000 {
/* x < 2**-25 */
return -TPI / x;
}
z = x * x;
u = U0[0] + z * (U0[1] + z * (U0[2] + z * (U0[3] + z * U0[4])));
v = 1.0 + z * (V0[0] + z * (V0[1] + z * (V0[2] + z * (V0[3] + z * V0[4]))));
return x * (u / v) + TPI * (j1f(x) * logf(x) - 1.0 / x);
}
/* For x >= 8, the asymptotic expansions of pone is
* 1 + 15/128 s^2 - 4725/2^15 s^4 - ..., where s = 1/x.
* We approximate pone by
* pone(x) = 1 + (R/S)
* where R = pr0 + pr1*s^2 + pr2*s^4 + ... + pr5*s^10
* S = 1 + ps0*s^2 + ... + ps4*s^10
* and
* | pone(x)-1-R/S | <= 2 ** ( -60.06)
*/
const PR8: [f32; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.0000000000e+00, /* 0x00000000 */
1.1718750000e-01, /* 0x3df00000 */
1.3239480972e+01, /* 0x4153d4ea */
4.1205184937e+02, /* 0x43ce06a3 */
3.8747453613e+03, /* 0x45722bed */
7.9144794922e+03, /* 0x45f753d6 */
];
const PS8: [f32; 5] = [
1.1420736694e+02, /* 0x42e46a2c */
3.6509309082e+03, /* 0x45642ee5 */
3.6956207031e+04, /* 0x47105c35 */
9.7602796875e+04, /* 0x47bea166 */
3.0804271484e+04, /* 0x46f0a88b */
];
const PR5: [f32; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
1.3199052094e-11, /* 0x2d68333f */
1.1718749255e-01, /* 0x3defffff */
6.8027510643e+00, /* 0x40d9b023 */
1.0830818176e+02, /* 0x42d89dca */
5.1763616943e+02, /* 0x440168b7 */
5.2871520996e+02, /* 0x44042dc6 */
];
const PS5: [f32; 5] = [
5.9280597687e+01, /* 0x426d1f55 */
9.9140142822e+02, /* 0x4477d9b1 */
5.3532670898e+03, /* 0x45a74a23 */
7.8446904297e+03, /* 0x45f52586 */
1.5040468750e+03, /* 0x44bc0180 */
];
const PR3: [f32; 6] = [
3.0250391081e-09, /* 0x314fe10d */
1.1718686670e-01, /* 0x3defffab */
3.9329774380e+00, /* 0x407bb5e7 */
3.5119403839e+01, /* 0x420c7a45 */
9.1055007935e+01, /* 0x42b61c2a */
4.8559066772e+01, /* 0x42423c7c */
];
const PS3: [f32; 5] = [
3.4791309357e+01, /* 0x420b2a4d */
3.3676245117e+02, /* 0x43a86198 */
1.0468714600e+03, /* 0x4482dbe3 */
8.9081134033e+02, /* 0x445eb3ed */
1.0378793335e+02, /* 0x42cf936c */
];
const PR2: [f32; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
1.0771083225e-07, /* 0x33e74ea8 */
1.1717621982e-01, /* 0x3deffa16 */
2.3685150146e+00, /* 0x401795c0 */
1.2242610931e+01, /* 0x4143e1bc */
1.7693971634e+01, /* 0x418d8d41 */
5.0735230446e+00, /* 0x40a25a4d */
];
const PS2: [f32; 5] = [
2.1436485291e+01, /* 0x41ab7dec */
1.2529022980e+02, /* 0x42fa9499 */
2.3227647400e+02, /* 0x436846c7 */
1.1767937469e+02, /* 0x42eb5bd7 */
8.3646392822e+00, /* 0x4105d590 */
];
fn ponef(x: f32) -> f32 {
let p: &[f32; 6];
let q: &[f32; 5];
let z: f32;
let r: f32;
let s: f32;
let mut ix: u32;
ix = x.to_bits();
ix &= 0x7fffffff;
if ix >= 0x41000000 {
p = &PR8;
q = &PS8;
} else if ix >= 0x409173eb {
p = &PR5;
q = &PS5;
} else if ix >= 0x4036d917 {
p = &PR3;
q = &PS3;
} else
/*ix >= 0x40000000*/
{
p = &PR2;
q = &PS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * q[4]))));
return 1.0 + r / s;
}
/* For x >= 8, the asymptotic expansions of qone is
* 3/8 s - 105/1024 s^3 - ..., where s = 1/x.
* We approximate pone by
* qone(x) = s*(0.375 + (R/S))
* where R = qr1*s^2 + qr2*s^4 + ... + qr5*s^10
* S = 1 + qs1*s^2 + ... + qs6*s^12
* and
* | qone(x)/s -0.375-R/S | <= 2 ** ( -61.13)
*/
const QR8: [f32; 6] = [
/* for x in [inf, 8]=1/[0,0.125] */
0.0000000000e+00, /* 0x00000000 */
-1.0253906250e-01, /* 0xbdd20000 */
-1.6271753311e+01, /* 0xc1822c8d */
-7.5960174561e+02, /* 0xc43de683 */
-1.1849806641e+04, /* 0xc639273a */
-4.8438511719e+04, /* 0xc73d3683 */
];
const QS8: [f32; 6] = [
1.6139537048e+02, /* 0x43216537 */
7.8253862305e+03, /* 0x45f48b17 */
1.3387534375e+05, /* 0x4802bcd6 */
7.1965775000e+05, /* 0x492fb29c */
6.6660125000e+05, /* 0x4922be94 */
-2.9449025000e+05, /* 0xc88fcb48 */
];
const QR5: [f32; 6] = [
/* for x in [8,4.5454]=1/[0.125,0.22001] */
-2.0897993405e-11, /* 0xadb7d219 */
-1.0253904760e-01, /* 0xbdd1fffe */
-8.0564479828e+00, /* 0xc100e736 */
-1.8366960144e+02, /* 0xc337ab6b */
-1.3731937256e+03, /* 0xc4aba633 */
-2.6124443359e+03, /* 0xc523471c */
];
const QS5: [f32; 6] = [
8.1276550293e+01, /* 0x42a28d98 */
1.9917987061e+03, /* 0x44f8f98f */
1.7468484375e+04, /* 0x468878f8 */
4.9851425781e+04, /* 0x4742bb6d */
2.7948074219e+04, /* 0x46da5826 */
-4.7191835938e+03, /* 0xc5937978 */
];
const QR3: [f32; 6] = [
-5.0783124372e-09, /* 0xb1ae7d4f */
-1.0253783315e-01, /* 0xbdd1ff5b */
-4.6101160049e+00, /* 0xc0938612 */
-5.7847221375e+01, /* 0xc267638e */
-2.2824453735e+02, /* 0xc3643e9a */
-2.1921012878e+02, /* 0xc35b35cb */
];
const QS3: [f32; 6] = [
4.7665153503e+01, /* 0x423ea91e */
6.7386511230e+02, /* 0x4428775e */
3.3801528320e+03, /* 0x45534272 */
5.5477290039e+03, /* 0x45ad5dd5 */
1.9031191406e+03, /* 0x44ede3d0 */
-1.3520118713e+02, /* 0xc3073381 */
];
const QR2: [f32; 6] = [
/* for x in [2.8570,2]=1/[0.3499,0.5] */
-1.7838172539e-07, /* 0xb43f8932 */
-1.0251704603e-01, /* 0xbdd1f475 */
-2.7522056103e+00, /* 0xc0302423 */
-1.9663616180e+01, /* 0xc19d4f16 */
-4.2325313568e+01, /* 0xc2294d1f */
-2.1371921539e+01, /* 0xc1aaf9b2 */
];
const QS2: [f32; 6] = [
2.9533363342e+01, /* 0x41ec4454 */
2.5298155212e+02, /* 0x437cfb47 */
7.5750280762e+02, /* 0x443d602e */
7.3939318848e+02, /* 0x4438d92a */
1.5594900513e+02, /* 0x431bf2f2 */
-4.9594988823e+00, /* 0xc09eb437 */
];
fn qonef(x: f32) -> f32 {
let p: &[f32; 6];
let q: &[f32; 6];
let s: f32;
let r: f32;
let z: f32;
let mut ix: u32;
ix = x.to_bits();
ix &= 0x7fffffff;
if ix >= 0x41000000 {
p = &QR8;
q = &QS8;
} else if ix >= 0x409173eb {
p = &QR5;
q = &QS5;
} else if ix >= 0x4036d917 {
p = &QR3;
q = &QS3;
} else
/*ix >= 0x40000000*/
{
p = &QR2;
q = &QS2;
}
z = 1.0 / (x * x);
r = p[0] + z * (p[1] + z * (p[2] + z * (p[3] + z * (p[4] + z * p[5]))));
s = 1.0 + z * (q[0] + z * (q[1] + z * (q[2] + z * (q[3] + z * (q[4] + z * q[5])))));
return (0.375 + r / s) / x;
}
// PowerPC tests are failing on LLVM 13: https://github.com/rust-lang/rust/issues/88520
#[cfg(not(target_arch = "powerpc64"))]
#[cfg(test)]
mod tests {
use super::{j1f, y1f};
#[test]
fn test_j1f_2488() {
// 0x401F3E49
assert_eq!(j1f(2.4881766_f32), 0.49999475_f32);
}
#[test]
fn test_y1f_2002() {
//allow slightly different result on x87
let res = y1f(2.0000002_f32);
if cfg!(all(target_arch = "x86", not(target_feature = "sse2"))) && (res == -0.10703231_f32)
{
return;
}
assert_eq!(res, -0.10703229_f32);
}
}

View File

@ -0,0 +1,335 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_jn.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* jn(n, x), yn(n, x)
* floating point Bessel's function of the 1st and 2nd kind
* of order n
*
* Special cases:
* y0(0)=y1(0)=yn(n,0) = -inf with division by zero signal;
* y0(-ve)=y1(-ve)=yn(n,-ve) are NaN with invalid signal.
* Note 2. About jn(n,x), yn(n,x)
* For n=0, j0(x) is called,
* for n=1, j1(x) is called,
* for n<=x, forward recursion is used starting
* from values of j0(x) and j1(x).
* for n>x, a continued fraction approximation to
* j(n,x)/j(n-1,x) is evaluated and then backward
* recursion is used starting from a supposed value
* for j(n,x). The resulting value of j(0,x) is
* compared with the actual value to correct the
* supposed value of j(n,x).
*
* yn(n,x) is similar in all respects, except
* that forward recursion is used for all
* values of n>1.
*/
use super::{cos, fabs, get_high_word, get_low_word, j0, j1, log, sin, sqrt, y0, y1};
const INVSQRTPI: f64 = 5.64189583547756279280e-01; /* 0x3FE20DD7, 0x50429B6D */
pub fn jn(n: i32, mut x: f64) -> f64 {
let mut ix: u32;
let lx: u32;
let nm1: i32;
let mut i: i32;
let mut sign: bool;
let mut a: f64;
let mut b: f64;
let mut temp: f64;
ix = get_high_word(x);
lx = get_low_word(x);
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
// -lx == !lx + 1
if (ix | (lx | ((!lx).wrapping_add(1))) >> 31) > 0x7ff00000 {
/* nan */
return x;
}
/* J(-n,x) = (-1)^n * J(n, x), J(n, -x) = (-1)^n * J(n, x)
* Thus, J(-n,x) = J(n,-x)
*/
/* nm1 = |n|-1 is used instead of |n| to handle n==INT_MIN */
if n == 0 {
return j0(x);
}
if n < 0 {
nm1 = -(n + 1);
x = -x;
sign = !sign;
} else {
nm1 = n - 1;
}
if nm1 == 0 {
return j1(x);
}
sign &= (n & 1) != 0; /* even n: 0, odd n: signbit(x) */
x = fabs(x);
if (ix | lx) == 0 || ix == 0x7ff00000 {
/* if x is 0 or inf */
b = 0.0;
} else if (nm1 as f64) < x {
/* Safe to use J(n+1,x)=2n/x *J(n,x)-J(n-1,x) */
if ix >= 0x52d00000 {
/* x > 2**302 */
/* (x >> n**2)
* Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Let s=sin(x), c=cos(x),
* xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
*
* n sin(xn)*sqt2 cos(xn)*sqt2
* ----------------------------------
* 0 s-c c+s
* 1 -s-c -c+s
* 2 -s+c -c-s
* 3 s+c c-s
*/
temp = match nm1 & 3 {
0 => -cos(x) + sin(x),
1 => -cos(x) - sin(x),
2 => cos(x) - sin(x),
3 | _ => cos(x) + sin(x),
};
b = INVSQRTPI * temp / sqrt(x);
} else {
a = j0(x);
b = j1(x);
i = 0;
while i < nm1 {
i += 1;
temp = b;
b = b * (2.0 * (i as f64) / x) - a; /* avoid underflow */
a = temp;
}
}
} else {
if ix < 0x3e100000 {
/* x < 2**-29 */
/* x is tiny, return the first Taylor expansion of J(n,x)
* J(n,x) = 1/n!*(x/2)^n - ...
*/
if nm1 > 32 {
/* underflow */
b = 0.0;
} else {
temp = x * 0.5;
b = temp;
a = 1.0;
i = 2;
while i <= nm1 + 1 {
a *= i as f64; /* a = n! */
b *= temp; /* b = (x/2)^n */
i += 1;
}
b = b / a;
}
} else {
/* use backward recurrence */
/* x x^2 x^2
* J(n,x)/J(n-1,x) = ---- ------ ------ .....
* 2n - 2(n+1) - 2(n+2)
*
* 1 1 1
* (for large x) = ---- ------ ------ .....
* 2n 2(n+1) 2(n+2)
* -- - ------ - ------ -
* x x x
*
* Let w = 2n/x and h=2/x, then the above quotient
* is equal to the continued fraction:
* 1
* = -----------------------
* 1
* w - -----------------
* 1
* w+h - ---------
* w+2h - ...
*
* To determine how many terms needed, let
* Q(0) = w, Q(1) = w(w+h) - 1,
* Q(k) = (w+k*h)*Q(k-1) - Q(k-2),
* When Q(k) > 1e4 good for single
* When Q(k) > 1e9 good for double
* When Q(k) > 1e17 good for quadruple
*/
/* determine k */
let mut t: f64;
let mut q0: f64;
let mut q1: f64;
let mut w: f64;
let h: f64;
let mut z: f64;
let mut tmp: f64;
let nf: f64;
let mut k: i32;
nf = (nm1 as f64) + 1.0;
w = 2.0 * nf / x;
h = 2.0 / x;
z = w + h;
q0 = w;
q1 = w * z - 1.0;
k = 1;
while q1 < 1.0e9 {
k += 1;
z += h;
tmp = z * q1 - q0;
q0 = q1;
q1 = tmp;
}
t = 0.0;
i = k;
while i >= 0 {
t = 1.0 / (2.0 * ((i as f64) + nf) / x - t);
i -= 1;
}
a = t;
b = 1.0;
/* estimate log((2/x)^n*n!) = n*log(2/x)+n*ln(n)
* Hence, if n*(log(2n/x)) > ...
* single 8.8722839355e+01
* double 7.09782712893383973096e+02
* long double 1.1356523406294143949491931077970765006170e+04
* then recurrent value may overflow and the result is
* likely underflow to zero
*/
tmp = nf * log(fabs(w));
if tmp < 7.09782712893383973096e+02 {
i = nm1;
while i > 0 {
temp = b;
b = b * (2.0 * (i as f64)) / x - a;
a = temp;
i -= 1;
}
} else {
i = nm1;
while i > 0 {
temp = b;
b = b * (2.0 * (i as f64)) / x - a;
a = temp;
/* scale b to avoid spurious overflow */
let x1p500 = f64::from_bits(0x5f30000000000000); // 0x1p500 == 2^500
if b > x1p500 {
a /= b;
t /= b;
b = 1.0;
}
i -= 1;
}
}
z = j0(x);
w = j1(x);
if fabs(z) >= fabs(w) {
b = t * z / b;
} else {
b = t * w / a;
}
}
}
if sign { -b } else { b }
}
pub fn yn(n: i32, x: f64) -> f64 {
let mut ix: u32;
let lx: u32;
let mut ib: u32;
let nm1: i32;
let mut sign: bool;
let mut i: i32;
let mut a: f64;
let mut b: f64;
let mut temp: f64;
ix = get_high_word(x);
lx = get_low_word(x);
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
// -lx == !lx + 1
if (ix | (lx | ((!lx).wrapping_add(1))) >> 31) > 0x7ff00000 {
/* nan */
return x;
}
if sign && (ix | lx) != 0 {
/* x < 0 */
return 0.0 / 0.0;
}
if ix == 0x7ff00000 {
return 0.0;
}
if n == 0 {
return y0(x);
}
if n < 0 {
nm1 = -(n + 1);
sign = (n & 1) != 0;
} else {
nm1 = n - 1;
sign = false;
}
if nm1 == 0 {
if sign {
return -y1(x);
} else {
return y1(x);
}
}
if ix >= 0x52d00000 {
/* x > 2**302 */
/* (x >> n**2)
* Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Let s=sin(x), c=cos(x),
* xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
*
* n sin(xn)*sqt2 cos(xn)*sqt2
* ----------------------------------
* 0 s-c c+s
* 1 -s-c -c+s
* 2 -s+c -c-s
* 3 s+c c-s
*/
temp = match nm1 & 3 {
0 => -sin(x) - cos(x),
1 => -sin(x) + cos(x),
2 => sin(x) + cos(x),
3 | _ => sin(x) - cos(x),
};
b = INVSQRTPI * temp / sqrt(x);
} else {
a = y0(x);
b = y1(x);
/* quit if b is -inf */
ib = get_high_word(b);
i = 0;
while i < nm1 && ib != 0xfff00000 {
i += 1;
temp = b;
b = (2.0 * (i as f64) / x) * b - a;
ib = get_high_word(b);
a = temp;
}
}
if sign { -b } else { b }
}

View File

@ -0,0 +1,251 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_jnf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{fabsf, j0f, j1f, logf, y0f, y1f};
pub fn jnf(n: i32, mut x: f32) -> f32 {
let mut ix: u32;
let mut nm1: i32;
let mut sign: bool;
let mut i: i32;
let mut a: f32;
let mut b: f32;
let mut temp: f32;
ix = x.to_bits();
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix > 0x7f800000 {
/* nan */
return x;
}
/* J(-n,x) = J(n,-x), use |n|-1 to avoid overflow in -n */
if n == 0 {
return j0f(x);
}
if n < 0 {
nm1 = -(n + 1);
x = -x;
sign = !sign;
} else {
nm1 = n - 1;
}
if nm1 == 0 {
return j1f(x);
}
sign &= (n & 1) != 0; /* even n: 0, odd n: signbit(x) */
x = fabsf(x);
if ix == 0 || ix == 0x7f800000 {
/* if x is 0 or inf */
b = 0.0;
} else if (nm1 as f32) < x {
/* Safe to use J(n+1,x)=2n/x *J(n,x)-J(n-1,x) */
a = j0f(x);
b = j1f(x);
i = 0;
while i < nm1 {
i += 1;
temp = b;
b = b * (2.0 * (i as f32) / x) - a;
a = temp;
}
} else {
if ix < 0x35800000 {
/* x < 2**-20 */
/* x is tiny, return the first Taylor expansion of J(n,x)
* J(n,x) = 1/n!*(x/2)^n - ...
*/
if nm1 > 8 {
/* underflow */
nm1 = 8;
}
temp = 0.5 * x;
b = temp;
a = 1.0;
i = 2;
while i <= nm1 + 1 {
a *= i as f32; /* a = n! */
b *= temp; /* b = (x/2)^n */
i += 1;
}
b = b / a;
} else {
/* use backward recurrence */
/* x x^2 x^2
* J(n,x)/J(n-1,x) = ---- ------ ------ .....
* 2n - 2(n+1) - 2(n+2)
*
* 1 1 1
* (for large x) = ---- ------ ------ .....
* 2n 2(n+1) 2(n+2)
* -- - ------ - ------ -
* x x x
*
* Let w = 2n/x and h=2/x, then the above quotient
* is equal to the continued fraction:
* 1
* = -----------------------
* 1
* w - -----------------
* 1
* w+h - ---------
* w+2h - ...
*
* To determine how many terms needed, let
* Q(0) = w, Q(1) = w(w+h) - 1,
* Q(k) = (w+k*h)*Q(k-1) - Q(k-2),
* When Q(k) > 1e4 good for single
* When Q(k) > 1e9 good for double
* When Q(k) > 1e17 good for quadruple
*/
/* determine k */
let mut t: f32;
let mut q0: f32;
let mut q1: f32;
let mut w: f32;
let h: f32;
let mut z: f32;
let mut tmp: f32;
let nf: f32;
let mut k: i32;
nf = (nm1 as f32) + 1.0;
w = 2.0 * (nf as f32) / x;
h = 2.0 / x;
z = w + h;
q0 = w;
q1 = w * z - 1.0;
k = 1;
while q1 < 1.0e4 {
k += 1;
z += h;
tmp = z * q1 - q0;
q0 = q1;
q1 = tmp;
}
t = 0.0;
i = k;
while i >= 0 {
t = 1.0 / (2.0 * ((i as f32) + nf) / x - t);
i -= 1;
}
a = t;
b = 1.0;
/* estimate log((2/x)^n*n!) = n*log(2/x)+n*ln(n)
* Hence, if n*(log(2n/x)) > ...
* single 8.8722839355e+01
* double 7.09782712893383973096e+02
* long double 1.1356523406294143949491931077970765006170e+04
* then recurrent value may overflow and the result is
* likely underflow to zero
*/
tmp = nf * logf(fabsf(w));
if tmp < 88.721679688 {
i = nm1;
while i > 0 {
temp = b;
b = 2.0 * (i as f32) * b / x - a;
a = temp;
i -= 1;
}
} else {
i = nm1;
while i > 0 {
temp = b;
b = 2.0 * (i as f32) * b / x - a;
a = temp;
/* scale b to avoid spurious overflow */
let x1p60 = f32::from_bits(0x5d800000); // 0x1p60 == 2^60
if b > x1p60 {
a /= b;
t /= b;
b = 1.0;
}
i -= 1;
}
}
z = j0f(x);
w = j1f(x);
if fabsf(z) >= fabsf(w) {
b = t * z / b;
} else {
b = t * w / a;
}
}
}
if sign { -b } else { b }
}
pub fn ynf(n: i32, x: f32) -> f32 {
let mut ix: u32;
let mut ib: u32;
let nm1: i32;
let mut sign: bool;
let mut i: i32;
let mut a: f32;
let mut b: f32;
let mut temp: f32;
ix = x.to_bits();
sign = (ix >> 31) != 0;
ix &= 0x7fffffff;
if ix > 0x7f800000 {
/* nan */
return x;
}
if sign && ix != 0 {
/* x < 0 */
return 0.0 / 0.0;
}
if ix == 0x7f800000 {
return 0.0;
}
if n == 0 {
return y0f(x);
}
if n < 0 {
nm1 = -(n + 1);
sign = (n & 1) != 0;
} else {
nm1 = n - 1;
sign = false;
}
if nm1 == 0 {
if sign {
return -y1f(x);
} else {
return y1f(x);
}
}
a = y0f(x);
b = y1f(x);
/* quit if b is -inf */
ib = b.to_bits();
i = 0;
while i < nm1 && ib != 0xff800000 {
i += 1;
temp = b;
b = (2.0 * (i as f32) / x) * b - a;
ib = b.to_bits();
a = temp;
}
if sign { -b } else { b }
}

View File

@ -0,0 +1,62 @@
// origin: FreeBSD /usr/src/lib/msun/src/k_cos.c
//
// ====================================================
// Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
//
// Developed at SunSoft, a Sun Microsystems, Inc. business.
// Permission to use, copy, modify, and distribute this
// software is freely granted, provided that this notice
// is preserved.
// ====================================================
const C1: f64 = 4.16666666666666019037e-02; /* 0x3FA55555, 0x5555554C */
const C2: f64 = -1.38888888888741095749e-03; /* 0xBF56C16C, 0x16C15177 */
const C3: f64 = 2.48015872894767294178e-05; /* 0x3EFA01A0, 0x19CB1590 */
const C4: f64 = -2.75573143513906633035e-07; /* 0xBE927E4F, 0x809C52AD */
const C5: f64 = 2.08757232129817482790e-09; /* 0x3E21EE9E, 0xBDB4B1C4 */
const C6: f64 = -1.13596475577881948265e-11; /* 0xBDA8FAE9, 0xBE8838D4 */
// kernel cos function on [-pi/4, pi/4], pi/4 ~ 0.785398164
// Input x is assumed to be bounded by ~pi/4 in magnitude.
// Input y is the tail of x.
//
// Algorithm
// 1. Since cos(-x) = cos(x), we need only to consider positive x.
// 2. if x < 2^-27 (hx<0x3e400000 0), return 1 with inexact if x!=0.
// 3. cos(x) is approximated by a polynomial of degree 14 on
// [0,pi/4]
// 4 14
// cos(x) ~ 1 - x*x/2 + C1*x + ... + C6*x
// where the remez error is
//
// | 2 4 6 8 10 12 14 | -58
// |cos(x)-(1-.5*x +C1*x +C2*x +C3*x +C4*x +C5*x +C6*x )| <= 2
// | |
//
// 4 6 8 10 12 14
// 4. let r = C1*x +C2*x +C3*x +C4*x +C5*x +C6*x , then
// cos(x) ~ 1 - x*x/2 + r
// since cos(x+y) ~ cos(x) - sin(x)*y
// ~ cos(x) - x*y,
// a correction term is necessary in cos(x) and hence
// cos(x+y) = 1 - (x*x/2 - (r - x*y))
// For better accuracy, rearrange to
// cos(x+y) ~ w + (tmp + (r-x*y))
// where w = 1 - x*x/2 and tmp is a tiny correction term
// (1 - x*x/2 == w + tmp exactly in infinite precision).
// The exactness of w + tmp in infinite precision depends on w
// and tmp having the same precision as x. If they have extra
// precision due to compiler bugs, then the extra precision is
// only good provided it is retained in all terms of the final
// expression for cos(). Retention happens in all cases tested
// under FreeBSD, so don't pessimize things by forcibly clipping
// any extra precision in w.
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_cos(x: f64, y: f64) -> f64 {
let z = x * x;
let w = z * z;
let r = z * (C1 + z * (C2 + z * C3)) + w * w * (C4 + z * (C5 + z * C6));
let hz = 0.5 * z;
let w = 1.0 - hz;
w + (((1.0 - w) - hz) + (z * r - x * y))
}

View File

@ -0,0 +1,29 @@
/* origin: FreeBSD /usr/src/lib/msun/src/k_cosf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
* Debugged and optimized by Bruce D. Evans.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* |cos(x) - c(x)| < 2**-34.1 (~[-5.37e-11, 5.295e-11]). */
const C0: f64 = -0.499999997251031003120; /* -0x1ffffffd0c5e81.0p-54 */
const C1: f64 = 0.0416666233237390631894; /* 0x155553e1053a42.0p-57 */
const C2: f64 = -0.00138867637746099294692; /* -0x16c087e80f1e27.0p-62 */
const C3: f64 = 0.0000243904487962774090654; /* 0x199342e0ee5069.0p-68 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_cosf(x: f64) -> f32 {
let z = x * x;
let w = z * z;
let r = C2 + z * C3;
(((1.0 + z * C0) + w * C1) + (w * z) * r) as f32
}

View File

@ -0,0 +1,14 @@
use super::exp;
/* k is such that k*ln2 has minimal relative error and x - kln2 > log(FLT_MIN) */
const K: i32 = 2043;
/* expf(x)/2 for x >= log(FLT_MAX), slightly better than 0.5f*expf(x/2)*expf(x/2) */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_expo2(x: f64) -> f64 {
let k_ln2 = f64::from_bits(0x40962066151add8b);
/* note that k is odd and scale*scale overflows */
let scale = f64::from_bits(((((0x3ff + K / 2) as u32) << 20) as u64) << 32);
/* exp(x - k ln2) * 2**(k-1) */
exp(x - k_ln2) * scale * scale
}

View File

@ -0,0 +1,14 @@
use super::expf;
/* k is such that k*ln2 has minimal relative error and x - kln2 > log(FLT_MIN) */
const K: i32 = 235;
/* expf(x)/2 for x >= log(FLT_MAX), slightly better than 0.5f*expf(x/2)*expf(x/2) */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_expo2f(x: f32) -> f32 {
let k_ln2 = f32::from_bits(0x4322e3bc);
/* note that k is odd and scale*scale overflows */
let scale = f32::from_bits(((0x7f + K / 2) as u32) << 23);
/* exp(x - k ln2) * 2**(k-1) */
expf(x - k_ln2) * scale * scale
}

View File

@ -0,0 +1,53 @@
// origin: FreeBSD /usr/src/lib/msun/src/k_sin.c
//
// ====================================================
// Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
//
// Developed at SunSoft, a Sun Microsystems, Inc. business.
// Permission to use, copy, modify, and distribute this
// software is freely granted, provided that this notice
// is preserved.
// ====================================================
const S1: f64 = -1.66666666666666324348e-01; /* 0xBFC55555, 0x55555549 */
const S2: f64 = 8.33333333332248946124e-03; /* 0x3F811111, 0x1110F8A6 */
const S3: f64 = -1.98412698298579493134e-04; /* 0xBF2A01A0, 0x19C161D5 */
const S4: f64 = 2.75573137070700676789e-06; /* 0x3EC71DE3, 0x57B1FE7D */
const S5: f64 = -2.50507602534068634195e-08; /* 0xBE5AE5E6, 0x8A2B9CEB */
const S6: f64 = 1.58969099521155010221e-10; /* 0x3DE5D93A, 0x5ACFD57C */
// kernel sin function on ~[-pi/4, pi/4] (except on -0), pi/4 ~ 0.7854
// Input x is assumed to be bounded by ~pi/4 in magnitude.
// Input y is the tail of x.
// Input iy indicates whether y is 0. (if iy=0, y assume to be 0).
//
// Algorithm
// 1. Since sin(-x) = -sin(x), we need only to consider positive x.
// 2. Callers must return sin(-0) = -0 without calling here since our
// odd polynomial is not evaluated in a way that preserves -0.
// Callers may do the optimization sin(x) ~ x for tiny x.
// 3. sin(x) is approximated by a polynomial of degree 13 on
// [0,pi/4]
// 3 13
// sin(x) ~ x + S1*x + ... + S6*x
// where
//
// |sin(x) 2 4 6 8 10 12 | -58
// |----- - (1+S1*x +S2*x +S3*x +S4*x +S5*x +S6*x )| <= 2
// | x |
//
// 4. sin(x+y) = sin(x) + sin'(x')*y
// ~ sin(x) + (1-x*x/2)*y
// For better accuracy, let
// 3 2 2 2 2
// r = x *(S2+x *(S3+x *(S4+x *(S5+x *S6))))
// then 3 2
// sin(x) = x + (S1*x + (x *(r-y/2)+y))
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_sin(x: f64, y: f64, iy: i32) -> f64 {
let z = x * x;
let w = z * z;
let r = S2 + z * (S3 + z * S4) + z * w * (S5 + z * S6);
let v = z * x;
if iy == 0 { x + v * (S1 + z * r) } else { x - ((z * (0.5 * y - v * r) - y) - v * S1) }
}

View File

@ -0,0 +1,30 @@
/* origin: FreeBSD /usr/src/lib/msun/src/k_sinf.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
* Optimized by Bruce D. Evans.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* |sin(x)/x - s(x)| < 2**-37.5 (~[-4.89e-12, 4.824e-12]). */
const S1: f64 = -0.166666666416265235595; /* -0x15555554cbac77.0p-55 */
const S2: f64 = 0.0083333293858894631756; /* 0x111110896efbb2.0p-59 */
const S3: f64 = -0.000198393348360966317347; /* -0x1a00f9e2cae774.0p-65 */
const S4: f64 = 0.0000027183114939898219064; /* 0x16cd878c3b46a7.0p-71 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_sinf(x: f64) -> f32 {
let z = x * x;
let w = z * z;
let r = S3 + z * S4;
let s = z * x;
((x + s * (S1 + z * S2)) + s * w * r) as f32
}

View File

@ -0,0 +1,105 @@
// origin: FreeBSD /usr/src/lib/msun/src/k_tan.c */
//
// ====================================================
// Copyright 2004 Sun Microsystems, Inc. All Rights Reserved.
//
// Permission to use, copy, modify, and distribute this
// software is freely granted, provided that this notice
// is preserved.
// ====================================================
// kernel tan function on ~[-pi/4, pi/4] (except on -0), pi/4 ~ 0.7854
// Input x is assumed to be bounded by ~pi/4 in magnitude.
// Input y is the tail of x.
// Input odd indicates whether tan (if odd = 0) or -1/tan (if odd = 1) is returned.
//
// Algorithm
// 1. Since tan(-x) = -tan(x), we need only to consider positive x.
// 2. Callers must return tan(-0) = -0 without calling here since our
// odd polynomial is not evaluated in a way that preserves -0.
// Callers may do the optimization tan(x) ~ x for tiny x.
// 3. tan(x) is approximated by a odd polynomial of degree 27 on
// [0,0.67434]
// 3 27
// tan(x) ~ x + T1*x + ... + T13*x
// where
//
// |tan(x) 2 4 26 | -59.2
// |----- - (1+T1*x +T2*x +.... +T13*x )| <= 2
// | x |
//
// Note: tan(x+y) = tan(x) + tan'(x)*y
// ~ tan(x) + (1+x*x)*y
// Therefore, for better accuracy in computing tan(x+y), let
// 3 2 2 2 2
// r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
// then
// 3 2
// tan(x+y) = x + (T1*x + (x *(r+y)+y))
//
// 4. For x in [0.67434,pi/4], let y = pi/4 - x, then
// tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
// = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
static T: [f64; 13] = [
3.33333333333334091986e-01, /* 3FD55555, 55555563 */
1.33333333333201242699e-01, /* 3FC11111, 1110FE7A */
5.39682539762260521377e-02, /* 3FABA1BA, 1BB341FE */
2.18694882948595424599e-02, /* 3F9664F4, 8406D637 */
8.86323982359930005737e-03, /* 3F8226E3, E96E8493 */
3.59207910759131235356e-03, /* 3F6D6D22, C9560328 */
1.45620945432529025516e-03, /* 3F57DBC8, FEE08315 */
5.88041240820264096874e-04, /* 3F4344D8, F2F26501 */
2.46463134818469906812e-04, /* 3F3026F7, 1A8D1068 */
7.81794442939557092300e-05, /* 3F147E88, A03792A6 */
7.14072491382608190305e-05, /* 3F12B80F, 32F0A7E9 */
-1.85586374855275456654e-05, /* BEF375CB, DB605373 */
2.59073051863633712884e-05, /* 3EFB2A70, 74BF7AD4 */
];
const PIO4: f64 = 7.85398163397448278999e-01; /* 3FE921FB, 54442D18 */
const PIO4_LO: f64 = 3.06161699786838301793e-17; /* 3C81A626, 33145C07 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_tan(mut x: f64, mut y: f64, odd: i32) -> f64 {
let hx = (f64::to_bits(x) >> 32) as u32;
let big = (hx & 0x7fffffff) >= 0x3FE59428; /* |x| >= 0.6744 */
if big {
let sign = hx >> 31;
if sign != 0 {
x = -x;
y = -y;
}
x = (PIO4 - x) + (PIO4_LO - y);
y = 0.0;
}
let z = x * x;
let w = z * z;
/*
* Break x^5*(T[1]+x^2*T[2]+...) into
* x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
* x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
*/
let r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] + w * T[11]))));
let v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] + w * T[12])))));
let s = z * x;
let r = y + z * (s * (r + v) + y) + s * T[0];
let w = x + r;
if big {
let sign = hx >> 31;
let s = 1.0 - 2.0 * odd as f64;
let v = s - 2.0 * (x + (r - w * w / (w + s)));
return if sign != 0 { -v } else { v };
}
if odd == 0 {
return w;
}
/* -1.0/(x+r) has up to 2ulp error, so compute it accurately */
let w0 = zero_low_word(w);
let v = r - (w0 - x); /* w0+v = r+x */
let a = -1.0 / w;
let a0 = zero_low_word(a);
a0 + a * (1.0 + a0 * w0 + a0 * v)
}
fn zero_low_word(x: f64) -> f64 {
f64::from_bits(f64::to_bits(x) & 0xFFFF_FFFF_0000_0000)
}

View File

@ -0,0 +1,46 @@
/* origin: FreeBSD /usr/src/lib/msun/src/k_tan.c */
/*
* ====================================================
* Copyright 2004 Sun Microsystems, Inc. All Rights Reserved.
*
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* |tan(x)/x - t(x)| < 2**-25.5 (~[-2e-08, 2e-08]). */
const T: [f64; 6] = [
0.333331395030791399758, /* 0x15554d3418c99f.0p-54 */
0.133392002712976742718, /* 0x1112fd38999f72.0p-55 */
0.0533812378445670393523, /* 0x1b54c91d865afe.0p-57 */
0.0245283181166547278873, /* 0x191df3908c33ce.0p-58 */
0.00297435743359967304927, /* 0x185dadfcecf44e.0p-61 */
0.00946564784943673166728, /* 0x1362b9bf971bcd.0p-59 */
];
#[cfg_attr(all(test), no_panic::no_panic)]
pub(crate) fn k_tanf(x: f64, odd: bool) -> f32 {
let z = x * x;
/*
* Split up the polynomial into small independent terms to give
* opportunities for parallel evaluation. The chosen splitting is
* micro-optimized for Athlons (XP, X64). It costs 2 multiplications
* relative to Horner's method on sequential machines.
*
* We add the small terms from lowest degree up for efficiency on
* non-sequential machines (the lowest degree terms tend to be ready
* earlier). Apart from this, we don't care about order of
* operations, and don't need to to care since we have precision to
* spare. However, the chosen splitting is good for accuracy too,
* and would give results as accurate as Horner's method if the
* small terms were added from highest degree down.
*/
let mut r = T[4] + z * T[5];
let t = T[2] + z * T[3];
let w = z * z;
let s = z * x;
let u = T[0] + z * T[1];
r = (x + s * u) + (s * w) * (t + w * r);
(if odd { -1. / r } else { r }) as f32
}

View File

@ -0,0 +1,4 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ldexp(x: f64, n: i32) -> f64 {
super::scalbn(x, n)
}

View File

@ -0,0 +1,4 @@
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn ldexpf(x: f32, n: i32) -> f32 {
super::scalbnf(x, n)
}

View File

@ -0,0 +1,6 @@
use super::lgamma_r;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn lgamma(x: f64) -> f64 {
lgamma_r(x).0
}

View File

@ -0,0 +1,320 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_lgamma_r.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*
*/
/* lgamma_r(x, signgamp)
* Reentrant version of the logarithm of the Gamma function
* with user provide pointer for the sign of Gamma(x).
*
* Method:
* 1. Argument Reduction for 0 < x <= 8
* Since gamma(1+s)=s*gamma(s), for x in [0,8], we may
* reduce x to a number in [1.5,2.5] by
* lgamma(1+s) = log(s) + lgamma(s)
* for example,
* lgamma(7.3) = log(6.3) + lgamma(6.3)
* = log(6.3*5.3) + lgamma(5.3)
* = log(6.3*5.3*4.3*3.3*2.3) + lgamma(2.3)
* 2. Polynomial approximation of lgamma around its
* minimun ymin=1.461632144968362245 to maintain monotonicity.
* On [ymin-0.23, ymin+0.27] (i.e., [1.23164,1.73163]), use
* Let z = x-ymin;
* lgamma(x) = -1.214862905358496078218 + z^2*poly(z)
* where
* poly(z) is a 14 degree polynomial.
* 2. Rational approximation in the primary interval [2,3]
* We use the following approximation:
* s = x-2.0;
* lgamma(x) = 0.5*s + s*P(s)/Q(s)
* with accuracy
* |P/Q - (lgamma(x)-0.5s)| < 2**-61.71
* Our algorithms are based on the following observation
*
* zeta(2)-1 2 zeta(3)-1 3
* lgamma(2+s) = s*(1-Euler) + --------- * s - --------- * s + ...
* 2 3
*
* where Euler = 0.5771... is the Euler constant, which is very
* close to 0.5.
*
* 3. For x>=8, we have
* lgamma(x)~(x-0.5)log(x)-x+0.5*log(2pi)+1/(12x)-1/(360x**3)+....
* (better formula:
* lgamma(x)~(x-0.5)*(log(x)-1)-.5*(log(2pi)-1) + ...)
* Let z = 1/x, then we approximation
* f(z) = lgamma(x) - (x-0.5)(log(x)-1)
* by
* 3 5 11
* w = w0 + w1*z + w2*z + w3*z + ... + w6*z
* where
* |w - f(z)| < 2**-58.74
*
* 4. For negative x, since (G is gamma function)
* -x*G(-x)*G(x) = PI/sin(PI*x),
* we have
* G(x) = PI/(sin(PI*x)*(-x)*G(-x))
* since G(-x) is positive, sign(G(x)) = sign(sin(PI*x)) for x<0
* Hence, for x<0, signgam = sign(sin(PI*x)) and
* lgamma(x) = log(|Gamma(x)|)
* = log(PI/(|x*sin(PI*x)|)) - lgamma(-x);
* Note: one should avoid compute PI*(-x) directly in the
* computation of sin(PI*(-x)).
*
* 5. Special Cases
* lgamma(2+s) ~ s*(1-Euler) for tiny s
* lgamma(1) = lgamma(2) = 0
* lgamma(x) ~ -log(|x|) for tiny x
* lgamma(0) = lgamma(neg.integer) = inf and raise divide-by-zero
* lgamma(inf) = inf
* lgamma(-inf) = inf (bug for bug compatible with C99!?)
*
*/
use super::{floor, k_cos, k_sin, log};
const PI: f64 = 3.14159265358979311600e+00; /* 0x400921FB, 0x54442D18 */
const A0: f64 = 7.72156649015328655494e-02; /* 0x3FB3C467, 0xE37DB0C8 */
const A1: f64 = 3.22467033424113591611e-01; /* 0x3FD4A34C, 0xC4A60FAD */
const A2: f64 = 6.73523010531292681824e-02; /* 0x3FB13E00, 0x1A5562A7 */
const A3: f64 = 2.05808084325167332806e-02; /* 0x3F951322, 0xAC92547B */
const A4: f64 = 7.38555086081402883957e-03; /* 0x3F7E404F, 0xB68FEFE8 */
const A5: f64 = 2.89051383673415629091e-03; /* 0x3F67ADD8, 0xCCB7926B */
const A6: f64 = 1.19270763183362067845e-03; /* 0x3F538A94, 0x116F3F5D */
const A7: f64 = 5.10069792153511336608e-04; /* 0x3F40B6C6, 0x89B99C00 */
const A8: f64 = 2.20862790713908385557e-04; /* 0x3F2CF2EC, 0xED10E54D */
const A9: f64 = 1.08011567247583939954e-04; /* 0x3F1C5088, 0x987DFB07 */
const A10: f64 = 2.52144565451257326939e-05; /* 0x3EFA7074, 0x428CFA52 */
const A11: f64 = 4.48640949618915160150e-05; /* 0x3F07858E, 0x90A45837 */
const TC: f64 = 1.46163214496836224576e+00; /* 0x3FF762D8, 0x6356BE3F */
const TF: f64 = -1.21486290535849611461e-01; /* 0xBFBF19B9, 0xBCC38A42 */
/* tt = -(tail of TF) */
const TT: f64 = -3.63867699703950536541e-18; /* 0xBC50C7CA, 0xA48A971F */
const T0: f64 = 4.83836122723810047042e-01; /* 0x3FDEF72B, 0xC8EE38A2 */
const T1: f64 = -1.47587722994593911752e-01; /* 0xBFC2E427, 0x8DC6C509 */
const T2: f64 = 6.46249402391333854778e-02; /* 0x3FB08B42, 0x94D5419B */
const T3: f64 = -3.27885410759859649565e-02; /* 0xBFA0C9A8, 0xDF35B713 */
const T4: f64 = 1.79706750811820387126e-02; /* 0x3F9266E7, 0x970AF9EC */
const T5: f64 = -1.03142241298341437450e-02; /* 0xBF851F9F, 0xBA91EC6A */
const T6: f64 = 6.10053870246291332635e-03; /* 0x3F78FCE0, 0xE370E344 */
const T7: f64 = -3.68452016781138256760e-03; /* 0xBF6E2EFF, 0xB3E914D7 */
const T8: f64 = 2.25964780900612472250e-03; /* 0x3F6282D3, 0x2E15C915 */
const T9: f64 = -1.40346469989232843813e-03; /* 0xBF56FE8E, 0xBF2D1AF1 */
const T10: f64 = 8.81081882437654011382e-04; /* 0x3F4CDF0C, 0xEF61A8E9 */
const T11: f64 = -5.38595305356740546715e-04; /* 0xBF41A610, 0x9C73E0EC */
const T12: f64 = 3.15632070903625950361e-04; /* 0x3F34AF6D, 0x6C0EBBF7 */
const T13: f64 = -3.12754168375120860518e-04; /* 0xBF347F24, 0xECC38C38 */
const T14: f64 = 3.35529192635519073543e-04; /* 0x3F35FD3E, 0xE8C2D3F4 */
const U0: f64 = -7.72156649015328655494e-02; /* 0xBFB3C467, 0xE37DB0C8 */
const U1: f64 = 6.32827064025093366517e-01; /* 0x3FE4401E, 0x8B005DFF */
const U2: f64 = 1.45492250137234768737e+00; /* 0x3FF7475C, 0xD119BD6F */
const U3: f64 = 9.77717527963372745603e-01; /* 0x3FEF4976, 0x44EA8450 */
const U4: f64 = 2.28963728064692451092e-01; /* 0x3FCD4EAE, 0xF6010924 */
const U5: f64 = 1.33810918536787660377e-02; /* 0x3F8B678B, 0xBF2BAB09 */
const V1: f64 = 2.45597793713041134822e+00; /* 0x4003A5D7, 0xC2BD619C */
const V2: f64 = 2.12848976379893395361e+00; /* 0x40010725, 0xA42B18F5 */
const V3: f64 = 7.69285150456672783825e-01; /* 0x3FE89DFB, 0xE45050AF */
const V4: f64 = 1.04222645593369134254e-01; /* 0x3FBAAE55, 0xD6537C88 */
const V5: f64 = 3.21709242282423911810e-03; /* 0x3F6A5ABB, 0x57D0CF61 */
const S0: f64 = -7.72156649015328655494e-02; /* 0xBFB3C467, 0xE37DB0C8 */
const S1: f64 = 2.14982415960608852501e-01; /* 0x3FCB848B, 0x36E20878 */
const S2: f64 = 3.25778796408930981787e-01; /* 0x3FD4D98F, 0x4F139F59 */
const S3: f64 = 1.46350472652464452805e-01; /* 0x3FC2BB9C, 0xBEE5F2F7 */
const S4: f64 = 2.66422703033638609560e-02; /* 0x3F9B481C, 0x7E939961 */
const S5: f64 = 1.84028451407337715652e-03; /* 0x3F5E26B6, 0x7368F239 */
const S6: f64 = 3.19475326584100867617e-05; /* 0x3F00BFEC, 0xDD17E945 */
const R1: f64 = 1.39200533467621045958e+00; /* 0x3FF645A7, 0x62C4AB74 */
const R2: f64 = 7.21935547567138069525e-01; /* 0x3FE71A18, 0x93D3DCDC */
const R3: f64 = 1.71933865632803078993e-01; /* 0x3FC601ED, 0xCCFBDF27 */
const R4: f64 = 1.86459191715652901344e-02; /* 0x3F9317EA, 0x742ED475 */
const R5: f64 = 7.77942496381893596434e-04; /* 0x3F497DDA, 0xCA41A95B */
const R6: f64 = 7.32668430744625636189e-06; /* 0x3EDEBAF7, 0xA5B38140 */
const W0: f64 = 4.18938533204672725052e-01; /* 0x3FDACFE3, 0x90C97D69 */
const W1: f64 = 8.33333333333329678849e-02; /* 0x3FB55555, 0x5555553B */
const W2: f64 = -2.77777777728775536470e-03; /* 0xBF66C16C, 0x16B02E5C */
const W3: f64 = 7.93650558643019558500e-04; /* 0x3F4A019F, 0x98CF38B6 */
const W4: f64 = -5.95187557450339963135e-04; /* 0xBF4380CB, 0x8C0FE741 */
const W5: f64 = 8.36339918996282139126e-04; /* 0x3F4B67BA, 0x4CDAD5D1 */
const W6: f64 = -1.63092934096575273989e-03; /* 0xBF5AB89D, 0x0B9E43E4 */
/* sin(PI*x) assuming x > 2^-100, if sin(PI*x)==0 the sign is arbitrary */
fn sin_pi(mut x: f64) -> f64 {
let mut n: i32;
/* spurious inexact if odd int */
x = 2.0 * (x * 0.5 - floor(x * 0.5)); /* x mod 2.0 */
n = (x * 4.0) as i32;
n = div!(n + 1, 2);
x -= (n as f64) * 0.5;
x *= PI;
match n {
1 => k_cos(x, 0.0),
2 => k_sin(-x, 0.0, 0),
3 => -k_cos(x, 0.0),
0 | _ => k_sin(x, 0.0, 0),
}
}
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn lgamma_r(mut x: f64) -> (f64, i32) {
let u: u64 = x.to_bits();
let mut t: f64;
let y: f64;
let mut z: f64;
let nadj: f64;
let p: f64;
let p1: f64;
let p2: f64;
let p3: f64;
let q: f64;
let mut r: f64;
let w: f64;
let ix: u32;
let sign: bool;
let i: i32;
let mut signgam: i32;
/* purge off +-inf, NaN, +-0, tiny and negative arguments */
signgam = 1;
sign = (u >> 63) != 0;
ix = ((u >> 32) as u32) & 0x7fffffff;
if ix >= 0x7ff00000 {
return (x * x, signgam);
}
if ix < (0x3ff - 70) << 20 {
/* |x|<2**-70, return -log(|x|) */
if sign {
x = -x;
signgam = -1;
}
return (-log(x), signgam);
}
if sign {
x = -x;
t = sin_pi(x);
if t == 0.0 {
/* -integer */
return (1.0 / (x - x), signgam);
}
if t > 0.0 {
signgam = -1;
} else {
t = -t;
}
nadj = log(PI / (t * x));
} else {
nadj = 0.0;
}
/* purge off 1 and 2 */
if (ix == 0x3ff00000 || ix == 0x40000000) && (u & 0xffffffff) == 0 {
r = 0.0;
}
/* for x < 2.0 */
else if ix < 0x40000000 {
if ix <= 0x3feccccc {
/* lgamma(x) = lgamma(x+1)-log(x) */
r = -log(x);
if ix >= 0x3FE76944 {
y = 1.0 - x;
i = 0;
} else if ix >= 0x3FCDA661 {
y = x - (TC - 1.0);
i = 1;
} else {
y = x;
i = 2;
}
} else {
r = 0.0;
if ix >= 0x3FFBB4C3 {
/* [1.7316,2] */
y = 2.0 - x;
i = 0;
} else if ix >= 0x3FF3B4C4 {
/* [1.23,1.73] */
y = x - TC;
i = 1;
} else {
y = x - 1.0;
i = 2;
}
}
match i {
0 => {
z = y * y;
p1 = A0 + z * (A2 + z * (A4 + z * (A6 + z * (A8 + z * A10))));
p2 = z * (A1 + z * (A3 + z * (A5 + z * (A7 + z * (A9 + z * A11)))));
p = y * p1 + p2;
r += p - 0.5 * y;
}
1 => {
z = y * y;
w = z * y;
p1 = T0 + w * (T3 + w * (T6 + w * (T9 + w * T12))); /* parallel comp */
p2 = T1 + w * (T4 + w * (T7 + w * (T10 + w * T13)));
p3 = T2 + w * (T5 + w * (T8 + w * (T11 + w * T14)));
p = z * p1 - (TT - w * (p2 + y * p3));
r += TF + p;
}
2 => {
p1 = y * (U0 + y * (U1 + y * (U2 + y * (U3 + y * (U4 + y * U5)))));
p2 = 1.0 + y * (V1 + y * (V2 + y * (V3 + y * (V4 + y * V5))));
r += -0.5 * y + p1 / p2;
}
#[cfg(debug_assertions)]
_ => unreachable!(),
#[cfg(not(debug_assertions))]
_ => {}
}
} else if ix < 0x40200000 {
/* x < 8.0 */
i = x as i32;
y = x - (i as f64);
p = y * (S0 + y * (S1 + y * (S2 + y * (S3 + y * (S4 + y * (S5 + y * S6))))));
q = 1.0 + y * (R1 + y * (R2 + y * (R3 + y * (R4 + y * (R5 + y * R6)))));
r = 0.5 * y + p / q;
z = 1.0; /* lgamma(1+s) = log(s) + lgamma(s) */
// TODO: In C, this was implemented using switch jumps with fallthrough.
// Does this implementation have performance problems?
if i >= 7 {
z *= y + 6.0;
}
if i >= 6 {
z *= y + 5.0;
}
if i >= 5 {
z *= y + 4.0;
}
if i >= 4 {
z *= y + 3.0;
}
if i >= 3 {
z *= y + 2.0;
r += log(z);
}
} else if ix < 0x43900000 {
/* 8.0 <= x < 2**58 */
t = log(x);
z = 1.0 / x;
y = z * z;
w = W0 + z * (W1 + y * (W2 + y * (W3 + y * (W4 + y * (W5 + y * W6)))));
r = (x - 0.5) * (t - 1.0) + w;
} else {
/* 2**58 <= x <= inf */
r = x * (log(x) - 1.0);
}
if sign {
r = nadj - r;
}
return (r, signgam);
}

View File

@ -0,0 +1,6 @@
use super::lgammaf_r;
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn lgammaf(x: f32) -> f32 {
lgammaf_r(x).0
}

View File

@ -0,0 +1,255 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_lgammaf_r.c */
/*
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
*/
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use super::{floorf, k_cosf, k_sinf, logf};
const PI: f32 = 3.1415927410e+00; /* 0x40490fdb */
const A0: f32 = 7.7215664089e-02; /* 0x3d9e233f */
const A1: f32 = 3.2246702909e-01; /* 0x3ea51a66 */
const A2: f32 = 6.7352302372e-02; /* 0x3d89f001 */
const A3: f32 = 2.0580807701e-02; /* 0x3ca89915 */
const A4: f32 = 7.3855509982e-03; /* 0x3bf2027e */
const A5: f32 = 2.8905137442e-03; /* 0x3b3d6ec6 */
const A6: f32 = 1.1927076848e-03; /* 0x3a9c54a1 */
const A7: f32 = 5.1006977446e-04; /* 0x3a05b634 */
const A8: f32 = 2.2086278477e-04; /* 0x39679767 */
const A9: f32 = 1.0801156895e-04; /* 0x38e28445 */
const A10: f32 = 2.5214456400e-05; /* 0x37d383a2 */
const A11: f32 = 4.4864096708e-05; /* 0x383c2c75 */
const TC: f32 = 1.4616321325e+00; /* 0x3fbb16c3 */
const TF: f32 = -1.2148628384e-01; /* 0xbdf8cdcd */
/* TT = -(tail of TF) */
const TT: f32 = 6.6971006518e-09; /* 0x31e61c52 */
const T0: f32 = 4.8383611441e-01; /* 0x3ef7b95e */
const T1: f32 = -1.4758771658e-01; /* 0xbe17213c */
const T2: f32 = 6.4624942839e-02; /* 0x3d845a15 */
const T3: f32 = -3.2788541168e-02; /* 0xbd064d47 */
const T4: f32 = 1.7970675603e-02; /* 0x3c93373d */
const T5: f32 = -1.0314224288e-02; /* 0xbc28fcfe */
const T6: f32 = 6.1005386524e-03; /* 0x3bc7e707 */
const T7: f32 = -3.6845202558e-03; /* 0xbb7177fe */
const T8: f32 = 2.2596477065e-03; /* 0x3b141699 */
const T9: f32 = -1.4034647029e-03; /* 0xbab7f476 */
const T10: f32 = 8.8108185446e-04; /* 0x3a66f867 */
const T11: f32 = -5.3859531181e-04; /* 0xba0d3085 */
const T12: f32 = 3.1563205994e-04; /* 0x39a57b6b */
const T13: f32 = -3.1275415677e-04; /* 0xb9a3f927 */
const T14: f32 = 3.3552918467e-04; /* 0x39afe9f7 */
const U0: f32 = -7.7215664089e-02; /* 0xbd9e233f */
const U1: f32 = 6.3282704353e-01; /* 0x3f2200f4 */
const U2: f32 = 1.4549225569e+00; /* 0x3fba3ae7 */
const U3: f32 = 9.7771751881e-01; /* 0x3f7a4bb2 */
const U4: f32 = 2.2896373272e-01; /* 0x3e6a7578 */
const U5: f32 = 1.3381091878e-02; /* 0x3c5b3c5e */
const V1: f32 = 2.4559779167e+00; /* 0x401d2ebe */
const V2: f32 = 2.1284897327e+00; /* 0x4008392d */
const V3: f32 = 7.6928514242e-01; /* 0x3f44efdf */
const V4: f32 = 1.0422264785e-01; /* 0x3dd572af */
const V5: f32 = 3.2170924824e-03; /* 0x3b52d5db */
const S0: f32 = -7.7215664089e-02; /* 0xbd9e233f */
const S1: f32 = 2.1498242021e-01; /* 0x3e5c245a */
const S2: f32 = 3.2577878237e-01; /* 0x3ea6cc7a */
const S3: f32 = 1.4635047317e-01; /* 0x3e15dce6 */
const S4: f32 = 2.6642270386e-02; /* 0x3cda40e4 */
const S5: f32 = 1.8402845599e-03; /* 0x3af135b4 */
const S6: f32 = 3.1947532989e-05; /* 0x3805ff67 */
const R1: f32 = 1.3920053244e+00; /* 0x3fb22d3b */
const R2: f32 = 7.2193557024e-01; /* 0x3f38d0c5 */
const R3: f32 = 1.7193385959e-01; /* 0x3e300f6e */
const R4: f32 = 1.8645919859e-02; /* 0x3c98bf54 */
const R5: f32 = 7.7794247773e-04; /* 0x3a4beed6 */
const R6: f32 = 7.3266842264e-06; /* 0x36f5d7bd */
const W0: f32 = 4.1893854737e-01; /* 0x3ed67f1d */
const W1: f32 = 8.3333335817e-02; /* 0x3daaaaab */
const W2: f32 = -2.7777778450e-03; /* 0xbb360b61 */
const W3: f32 = 7.9365057172e-04; /* 0x3a500cfd */
const W4: f32 = -5.9518753551e-04; /* 0xba1c065c */
const W5: f32 = 8.3633989561e-04; /* 0x3a5b3dd2 */
const W6: f32 = -1.6309292987e-03; /* 0xbad5c4e8 */
/* sin(PI*x) assuming x > 2^-100, if sin(PI*x)==0 the sign is arbitrary */
fn sin_pi(mut x: f32) -> f32 {
let mut y: f64;
let mut n: isize;
/* spurious inexact if odd int */
x = 2.0 * (x * 0.5 - floorf(x * 0.5)); /* x mod 2.0 */
n = (x * 4.0) as isize;
n = div!(n + 1, 2);
y = (x as f64) - (n as f64) * 0.5;
y *= 3.14159265358979323846;
match n {
1 => k_cosf(y),
2 => k_sinf(-y),
3 => -k_cosf(y),
0 | _ => k_sinf(y),
}
}
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn lgammaf_r(mut x: f32) -> (f32, i32) {
let u = x.to_bits();
let mut t: f32;
let y: f32;
let mut z: f32;
let nadj: f32;
let p: f32;
let p1: f32;
let p2: f32;
let p3: f32;
let q: f32;
let mut r: f32;
let w: f32;
let ix: u32;
let i: i32;
let sign: bool;
let mut signgam: i32;
/* purge off +-inf, NaN, +-0, tiny and negative arguments */
signgam = 1;
sign = (u >> 31) != 0;
ix = u & 0x7fffffff;
if ix >= 0x7f800000 {
return (x * x, signgam);
}
if ix < 0x35000000 {
/* |x| < 2**-21, return -log(|x|) */
if sign {
signgam = -1;
x = -x;
}
return (-logf(x), signgam);
}
if sign {
x = -x;
t = sin_pi(x);
if t == 0.0 {
/* -integer */
return (1.0 / (x - x), signgam);
}
if t > 0.0 {
signgam = -1;
} else {
t = -t;
}
nadj = logf(PI / (t * x));
} else {
nadj = 0.0;
}
/* purge off 1 and 2 */
if ix == 0x3f800000 || ix == 0x40000000 {
r = 0.0;
}
/* for x < 2.0 */
else if ix < 0x40000000 {
if ix <= 0x3f666666 {
/* lgamma(x) = lgamma(x+1)-log(x) */
r = -logf(x);
if ix >= 0x3f3b4a20 {
y = 1.0 - x;
i = 0;
} else if ix >= 0x3e6d3308 {
y = x - (TC - 1.0);
i = 1;
} else {
y = x;
i = 2;
}
} else {
r = 0.0;
if ix >= 0x3fdda618 {
/* [1.7316,2] */
y = 2.0 - x;
i = 0;
} else if ix >= 0x3F9da620 {
/* [1.23,1.73] */
y = x - TC;
i = 1;
} else {
y = x - 1.0;
i = 2;
}
}
match i {
0 => {
z = y * y;
p1 = A0 + z * (A2 + z * (A4 + z * (A6 + z * (A8 + z * A10))));
p2 = z * (A1 + z * (A3 + z * (A5 + z * (A7 + z * (A9 + z * A11)))));
p = y * p1 + p2;
r += p - 0.5 * y;
}
1 => {
z = y * y;
w = z * y;
p1 = T0 + w * (T3 + w * (T6 + w * (T9 + w * T12))); /* parallel comp */
p2 = T1 + w * (T4 + w * (T7 + w * (T10 + w * T13)));
p3 = T2 + w * (T5 + w * (T8 + w * (T11 + w * T14)));
p = z * p1 - (TT - w * (p2 + y * p3));
r += TF + p;
}
2 => {
p1 = y * (U0 + y * (U1 + y * (U2 + y * (U3 + y * (U4 + y * U5)))));
p2 = 1.0 + y * (V1 + y * (V2 + y * (V3 + y * (V4 + y * V5))));
r += -0.5 * y + p1 / p2;
}
#[cfg(debug_assertions)]
_ => unreachable!(),
#[cfg(not(debug_assertions))]
_ => {}
}
} else if ix < 0x41000000 {
/* x < 8.0 */
i = x as i32;
y = x - (i as f32);
p = y * (S0 + y * (S1 + y * (S2 + y * (S3 + y * (S4 + y * (S5 + y * S6))))));
q = 1.0 + y * (R1 + y * (R2 + y * (R3 + y * (R4 + y * (R5 + y * R6)))));
r = 0.5 * y + p / q;
z = 1.0; /* lgamma(1+s) = log(s) + lgamma(s) */
// TODO: In C, this was implemented using switch jumps with fallthrough.
// Does this implementation have performance problems?
if i >= 7 {
z *= y + 6.0;
}
if i >= 6 {
z *= y + 5.0;
}
if i >= 5 {
z *= y + 4.0;
}
if i >= 4 {
z *= y + 3.0;
}
if i >= 3 {
z *= y + 2.0;
r += logf(z);
}
} else if ix < 0x5c800000 {
/* 8.0 <= x < 2**58 */
t = logf(x);
z = 1.0 / x;
y = z * z;
w = W0 + z * (W1 + y * (W2 + y * (W3 + y * (W4 + y * (W5 + y * W6)))));
r = (x - 0.5) * (t - 1.0) + w;
} else {
/* 2**58 <= x <= inf */
r = x * (logf(x) - 1.0);
}
if sign {
r = nadj - r;
}
return (r, signgam);
}

View File

@ -0,0 +1,117 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_log.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* log(x)
* Return the logarithm of x
*
* Method :
* 1. Argument Reduction: find k and f such that
* x = 2^k * (1+f),
* where sqrt(2)/2 < 1+f < sqrt(2) .
*
* 2. Approximation of log(1+f).
* Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s)
* = 2s + 2/3 s**3 + 2/5 s**5 + .....,
* = 2s + s*R
* We use a special Remez algorithm on [0,0.1716] to generate
* a polynomial of degree 14 to approximate R The maximum error
* of this polynomial approximation is bounded by 2**-58.45. In
* other words,
* 2 4 6 8 10 12 14
* R(z) ~ Lg1*s +Lg2*s +Lg3*s +Lg4*s +Lg5*s +Lg6*s +Lg7*s
* (the values of Lg1 to Lg7 are listed in the program)
* and
* | 2 14 | -58.45
* | Lg1*s +...+Lg7*s - R(z) | <= 2
* | |
* Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2.
* In order to guarantee error in log below 1ulp, we compute log
* by
* log(1+f) = f - s*(f - R) (if f is not too large)
* log(1+f) = f - (hfsq - s*(hfsq+R)). (better accuracy)
*
* 3. Finally, log(x) = k*ln2 + log(1+f).
* = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo)))
* Here ln2 is split into two floating point number:
* ln2_hi + ln2_lo,
* where n*ln2_hi is always exact for |n| < 2000.
*
* Special cases:
* log(x) is NaN with signal if x < 0 (including -INF) ;
* log(+INF) is +INF; log(0) is -INF with signal;
* log(NaN) is that NaN with no signal.
*
* Accuracy:
* according to an error analysis, the error is always less than
* 1 ulp (unit in the last place).
*
* Constants:
* The hexadecimal values are the intended ones for the following
* constants. The decimal values may be used, provided that the
* compiler will convert from decimal to binary accurately enough
* to produce the hexadecimal values shown.
*/
const LN2_HI: f64 = 6.93147180369123816490e-01; /* 3fe62e42 fee00000 */
const LN2_LO: f64 = 1.90821492927058770002e-10; /* 3dea39ef 35793c76 */
const LG1: f64 = 6.666666666666735130e-01; /* 3FE55555 55555593 */
const LG2: f64 = 3.999999999940941908e-01; /* 3FD99999 9997FA04 */
const LG3: f64 = 2.857142874366239149e-01; /* 3FD24924 94229359 */
const LG4: f64 = 2.222219843214978396e-01; /* 3FCC71C5 1D8E78AF */
const LG5: f64 = 1.818357216161805012e-01; /* 3FC74664 96CB03DE */
const LG6: f64 = 1.531383769920937332e-01; /* 3FC39A09 D078C69F */
const LG7: f64 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log(mut x: f64) -> f64 {
let x1p54 = f64::from_bits(0x4350000000000000); // 0x1p54 === 2 ^ 54
let mut ui = x.to_bits();
let mut hx: u32 = (ui >> 32) as u32;
let mut k: i32 = 0;
if (hx < 0x00100000) || ((hx >> 31) != 0) {
/* x < 2**-126 */
if ui << 1 == 0 {
return -1. / (x * x); /* log(+-0)=-inf */
}
if hx >> 31 != 0 {
return (x - x) / 0.0; /* log(-#) = NaN */
}
/* subnormal number, scale x up */
k -= 54;
x *= x1p54;
ui = x.to_bits();
hx = (ui >> 32) as u32;
} else if hx >= 0x7ff00000 {
return x;
} else if hx == 0x3ff00000 && ui << 32 == 0 {
return 0.;
}
/* reduce x into [sqrt(2)/2, sqrt(2)] */
hx += 0x3ff00000 - 0x3fe6a09e;
k += ((hx >> 20) as i32) - 0x3ff;
hx = (hx & 0x000fffff) + 0x3fe6a09e;
ui = ((hx as u64) << 32) | (ui & 0xffffffff);
x = f64::from_bits(ui);
let f: f64 = x - 1.0;
let hfsq: f64 = 0.5 * f * f;
let s: f64 = f / (2.0 + f);
let z: f64 = s * s;
let w: f64 = z * z;
let t1: f64 = w * (LG2 + w * (LG4 + w * LG6));
let t2: f64 = z * (LG1 + w * (LG3 + w * (LG5 + w * LG7)));
let r: f64 = t2 + t1;
let dk: f64 = k as f64;
s * (hfsq + r) + dk * LN2_LO - hfsq + f + dk * LN2_HI
}

View File

@ -0,0 +1,117 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_log10.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* Return the base 10 logarithm of x. See log.c for most comments.
*
* Reduce x to 2^k (1+f) and calculate r = log(1+f) - f + f*f/2
* as in log.c, then combine and scale in extra precision:
* log10(x) = (f - f*f/2 + r)/log(10) + k*log10(2)
*/
use core::f64;
const IVLN10HI: f64 = 4.34294481878168880939e-01; /* 0x3fdbcb7b, 0x15200000 */
const IVLN10LO: f64 = 2.50829467116452752298e-11; /* 0x3dbb9438, 0xca9aadd5 */
const LOG10_2HI: f64 = 3.01029995663611771306e-01; /* 0x3FD34413, 0x509F6000 */
const LOG10_2LO: f64 = 3.69423907715893078616e-13; /* 0x3D59FEF3, 0x11F12B36 */
const LG1: f64 = 6.666666666666735130e-01; /* 3FE55555 55555593 */
const LG2: f64 = 3.999999999940941908e-01; /* 3FD99999 9997FA04 */
const LG3: f64 = 2.857142874366239149e-01; /* 3FD24924 94229359 */
const LG4: f64 = 2.222219843214978396e-01; /* 3FCC71C5 1D8E78AF */
const LG5: f64 = 1.818357216161805012e-01; /* 3FC74664 96CB03DE */
const LG6: f64 = 1.531383769920937332e-01; /* 3FC39A09 D078C69F */
const LG7: f64 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log10(mut x: f64) -> f64 {
let x1p54 = f64::from_bits(0x4350000000000000); // 0x1p54 === 2 ^ 54
let mut ui: u64 = x.to_bits();
let hfsq: f64;
let f: f64;
let s: f64;
let z: f64;
let r: f64;
let mut w: f64;
let t1: f64;
let t2: f64;
let dk: f64;
let y: f64;
let mut hi: f64;
let lo: f64;
let mut val_hi: f64;
let mut val_lo: f64;
let mut hx: u32;
let mut k: i32;
hx = (ui >> 32) as u32;
k = 0;
if hx < 0x00100000 || (hx >> 31) > 0 {
if ui << 1 == 0 {
return -1. / (x * x); /* log(+-0)=-inf */
}
if (hx >> 31) > 0 {
return (x - x) / 0.0; /* log(-#) = NaN */
}
/* subnormal number, scale x up */
k -= 54;
x *= x1p54;
ui = x.to_bits();
hx = (ui >> 32) as u32;
} else if hx >= 0x7ff00000 {
return x;
} else if hx == 0x3ff00000 && ui << 32 == 0 {
return 0.;
}
/* reduce x into [sqrt(2)/2, sqrt(2)] */
hx += 0x3ff00000 - 0x3fe6a09e;
k += (hx >> 20) as i32 - 0x3ff;
hx = (hx & 0x000fffff) + 0x3fe6a09e;
ui = (hx as u64) << 32 | (ui & 0xffffffff);
x = f64::from_bits(ui);
f = x - 1.0;
hfsq = 0.5 * f * f;
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * (LG4 + w * LG6));
t2 = z * (LG1 + w * (LG3 + w * (LG5 + w * LG7)));
r = t2 + t1;
/* See log2.c for details. */
/* hi+lo = f - hfsq + s*(hfsq+R) ~ log(1+f) */
hi = f - hfsq;
ui = hi.to_bits();
ui &= (-1i64 as u64) << 32;
hi = f64::from_bits(ui);
lo = f - hi - hfsq + s * (hfsq + r);
/* val_hi+val_lo ~ log10(1+f) + k*log10(2) */
val_hi = hi * IVLN10HI;
dk = k as f64;
y = dk * LOG10_2HI;
val_lo = dk * LOG10_2LO + (lo + hi) * IVLN10LO + lo * IVLN10HI;
/*
* Extra precision in for adding y is not strictly needed
* since there is no very large cancellation near x = sqrt(2) or
* x = 1/sqrt(2), but we do it anyway since it costs little on CPUs
* with some parallelism and it reduces the error for many args.
*/
w = y + val_hi;
val_lo += (y - w) + val_hi;
val_hi = w;
val_lo + val_hi
}

View File

@ -0,0 +1,91 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_log10f.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* See comments in log10.c.
*/
use core::f32;
const IVLN10HI: f32 = 4.3432617188e-01; /* 0x3ede6000 */
const IVLN10LO: f32 = -3.1689971365e-05; /* 0xb804ead9 */
const LOG10_2HI: f32 = 3.0102920532e-01; /* 0x3e9a2080 */
const LOG10_2LO: f32 = 7.9034151668e-07; /* 0x355427db */
/* |(log(1+s)-log(1-s))/s - Lg(s)| < 2**-34.24 (~[-4.95e-11, 4.97e-11]). */
const LG1: f32 = 0.66666662693; /* 0xaaaaaa.0p-24 */
const LG2: f32 = 0.40000972152; /* 0xccce13.0p-25 */
const LG3: f32 = 0.28498786688; /* 0x91e9ee.0p-25 */
const LG4: f32 = 0.24279078841; /* 0xf89e26.0p-26 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log10f(mut x: f32) -> f32 {
let x1p25f = f32::from_bits(0x4c000000); // 0x1p25f === 2 ^ 25
let mut ui: u32 = x.to_bits();
let hfsq: f32;
let f: f32;
let s: f32;
let z: f32;
let r: f32;
let w: f32;
let t1: f32;
let t2: f32;
let dk: f32;
let mut hi: f32;
let lo: f32;
let mut ix: u32;
let mut k: i32;
ix = ui;
k = 0;
if ix < 0x00800000 || (ix >> 31) > 0 {
/* x < 2**-126 */
if ix << 1 == 0 {
return -1. / (x * x); /* log(+-0)=-inf */
}
if (ix >> 31) > 0 {
return (x - x) / 0.0; /* log(-#) = NaN */
}
/* subnormal number, scale up x */
k -= 25;
x *= x1p25f;
ui = x.to_bits();
ix = ui;
} else if ix >= 0x7f800000 {
return x;
} else if ix == 0x3f800000 {
return 0.;
}
/* reduce x into [sqrt(2)/2, sqrt(2)] */
ix += 0x3f800000 - 0x3f3504f3;
k += (ix >> 23) as i32 - 0x7f;
ix = (ix & 0x007fffff) + 0x3f3504f3;
ui = ix;
x = f32::from_bits(ui);
f = x - 1.0;
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * LG4);
t2 = z * (LG1 + w * LG3);
r = t2 + t1;
hfsq = 0.5 * f * f;
hi = f - hfsq;
ui = hi.to_bits();
ui &= 0xfffff000;
hi = f32::from_bits(ui);
lo = f - hi - hfsq + s * (hfsq + r);
dk = k as f32;
dk * LOG10_2LO + (lo + hi) * IVLN10LO + lo * IVLN10HI + hi * IVLN10HI + dk * LOG10_2HI
}

View File

@ -0,0 +1,139 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_log1p.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/* double log1p(double x)
* Return the natural logarithm of 1+x.
*
* Method :
* 1. Argument Reduction: find k and f such that
* 1+x = 2^k * (1+f),
* where sqrt(2)/2 < 1+f < sqrt(2) .
*
* Note. If k=0, then f=x is exact. However, if k!=0, then f
* may not be representable exactly. In that case, a correction
* term is need. Let u=1+x rounded. Let c = (1+x)-u, then
* log(1+x) - log(u) ~ c/u. Thus, we proceed to compute log(u),
* and add back the correction term c/u.
* (Note: when x > 2**53, one can simply return log(x))
*
* 2. Approximation of log(1+f): See log.c
*
* 3. Finally, log1p(x) = k*ln2 + log(1+f) + c/u. See log.c
*
* Special cases:
* log1p(x) is NaN with signal if x < -1 (including -INF) ;
* log1p(+INF) is +INF; log1p(-1) is -INF with signal;
* log1p(NaN) is that NaN with no signal.
*
* Accuracy:
* according to an error analysis, the error is always less than
* 1 ulp (unit in the last place).
*
* Constants:
* The hexadecimal values are the intended ones for the following
* constants. The decimal values may be used, provided that the
* compiler will convert from decimal to binary accurately enough
* to produce the hexadecimal values shown.
*
* Note: Assuming log() return accurate answer, the following
* algorithm can be used to compute log1p(x) to within a few ULP:
*
* u = 1+x;
* if(u==1.0) return x ; else
* return log(u)*(x/(u-1.0));
*
* See HP-15C Advanced Functions Handbook, p.193.
*/
use core::f64;
const LN2_HI: f64 = 6.93147180369123816490e-01; /* 3fe62e42 fee00000 */
const LN2_LO: f64 = 1.90821492927058770002e-10; /* 3dea39ef 35793c76 */
const LG1: f64 = 6.666666666666735130e-01; /* 3FE55555 55555593 */
const LG2: f64 = 3.999999999940941908e-01; /* 3FD99999 9997FA04 */
const LG3: f64 = 2.857142874366239149e-01; /* 3FD24924 94229359 */
const LG4: f64 = 2.222219843214978396e-01; /* 3FCC71C5 1D8E78AF */
const LG5: f64 = 1.818357216161805012e-01; /* 3FC74664 96CB03DE */
const LG6: f64 = 1.531383769920937332e-01; /* 3FC39A09 D078C69F */
const LG7: f64 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log1p(x: f64) -> f64 {
let mut ui: u64 = x.to_bits();
let hfsq: f64;
let mut f: f64 = 0.;
let mut c: f64 = 0.;
let s: f64;
let z: f64;
let r: f64;
let w: f64;
let t1: f64;
let t2: f64;
let dk: f64;
let hx: u32;
let mut hu: u32;
let mut k: i32;
hx = (ui >> 32) as u32;
k = 1;
if hx < 0x3fda827a || (hx >> 31) > 0 {
/* 1+x < sqrt(2)+ */
if hx >= 0xbff00000 {
/* x <= -1.0 */
if x == -1. {
return x / 0.0; /* log1p(-1) = -inf */
}
return (x - x) / 0.0; /* log1p(x<-1) = NaN */
}
if hx << 1 < 0x3ca00000 << 1 {
/* |x| < 2**-53 */
/* underflow if subnormal */
if (hx & 0x7ff00000) == 0 {
force_eval!(x as f32);
}
return x;
}
if hx <= 0xbfd2bec4 {
/* sqrt(2)/2- <= 1+x < sqrt(2)+ */
k = 0;
c = 0.;
f = x;
}
} else if hx >= 0x7ff00000 {
return x;
}
if k > 0 {
ui = (1. + x).to_bits();
hu = (ui >> 32) as u32;
hu += 0x3ff00000 - 0x3fe6a09e;
k = (hu >> 20) as i32 - 0x3ff;
/* correction term ~ log(1+x)-log(u), avoid underflow in c/u */
if k < 54 {
c = if k >= 2 { 1. - (f64::from_bits(ui) - x) } else { x - (f64::from_bits(ui) - 1.) };
c /= f64::from_bits(ui);
} else {
c = 0.;
}
/* reduce u into [sqrt(2)/2, sqrt(2)] */
hu = (hu & 0x000fffff) + 0x3fe6a09e;
ui = (hu as u64) << 32 | (ui & 0xffffffff);
f = f64::from_bits(ui) - 1.;
}
hfsq = 0.5 * f * f;
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * (LG4 + w * LG6));
t2 = z * (LG1 + w * (LG3 + w * (LG5 + w * LG7)));
r = t2 + t1;
dk = k as f64;
s * (hfsq + r) + (dk * LN2_LO + c) - hfsq + f + dk * LN2_HI
}

View File

@ -0,0 +1,94 @@
/* origin: FreeBSD /usr/src/lib/msun/src/s_log1pf.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
use core::f32;
const LN2_HI: f32 = 6.9313812256e-01; /* 0x3f317180 */
const LN2_LO: f32 = 9.0580006145e-06; /* 0x3717f7d1 */
/* |(log(1+s)-log(1-s))/s - Lg(s)| < 2**-34.24 (~[-4.95e-11, 4.97e-11]). */
const LG1: f32 = 0.66666662693; /* 0xaaaaaa.0p-24 */
const LG2: f32 = 0.40000972152; /* 0xccce13.0p-25 */
const LG3: f32 = 0.28498786688; /* 0x91e9ee.0p-25 */
const LG4: f32 = 0.24279078841; /* 0xf89e26.0p-26 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log1pf(x: f32) -> f32 {
let mut ui: u32 = x.to_bits();
let hfsq: f32;
let mut f: f32 = 0.;
let mut c: f32 = 0.;
let s: f32;
let z: f32;
let r: f32;
let w: f32;
let t1: f32;
let t2: f32;
let dk: f32;
let ix: u32;
let mut iu: u32;
let mut k: i32;
ix = ui;
k = 1;
if ix < 0x3ed413d0 || (ix >> 31) > 0 {
/* 1+x < sqrt(2)+ */
if ix >= 0xbf800000 {
/* x <= -1.0 */
if x == -1. {
return x / 0.0; /* log1p(-1)=+inf */
}
return (x - x) / 0.0; /* log1p(x<-1)=NaN */
}
if ix << 1 < 0x33800000 << 1 {
/* |x| < 2**-24 */
/* underflow if subnormal */
if (ix & 0x7f800000) == 0 {
force_eval!(x * x);
}
return x;
}
if ix <= 0xbe95f619 {
/* sqrt(2)/2- <= 1+x < sqrt(2)+ */
k = 0;
c = 0.;
f = x;
}
} else if ix >= 0x7f800000 {
return x;
}
if k > 0 {
ui = (1. + x).to_bits();
iu = ui;
iu += 0x3f800000 - 0x3f3504f3;
k = (iu >> 23) as i32 - 0x7f;
/* correction term ~ log(1+x)-log(u), avoid underflow in c/u */
if k < 25 {
c = if k >= 2 { 1. - (f32::from_bits(ui) - x) } else { x - (f32::from_bits(ui) - 1.) };
c /= f32::from_bits(ui);
} else {
c = 0.;
}
/* reduce u into [sqrt(2)/2, sqrt(2)] */
iu = (iu & 0x007fffff) + 0x3f3504f3;
ui = iu;
f = f32::from_bits(ui) - 1.;
}
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * LG4);
t2 = z * (LG1 + w * LG3);
r = t2 + t1;
hfsq = 0.5 * f * f;
dk = k as f32;
s * (hfsq + r) + (dk * LN2_LO + c) - hfsq + f + dk * LN2_HI
}

View File

@ -0,0 +1,106 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_log2.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunSoft, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* Return the base 2 logarithm of x. See log.c for most comments.
*
* Reduce x to 2^k (1+f) and calculate r = log(1+f) - f + f*f/2
* as in log.c, then combine and scale in extra precision:
* log2(x) = (f - f*f/2 + r)/log(2) + k
*/
use core::f64;
const IVLN2HI: f64 = 1.44269504072144627571e+00; /* 0x3ff71547, 0x65200000 */
const IVLN2LO: f64 = 1.67517131648865118353e-10; /* 0x3de705fc, 0x2eefa200 */
const LG1: f64 = 6.666666666666735130e-01; /* 3FE55555 55555593 */
const LG2: f64 = 3.999999999940941908e-01; /* 3FD99999 9997FA04 */
const LG3: f64 = 2.857142874366239149e-01; /* 3FD24924 94229359 */
const LG4: f64 = 2.222219843214978396e-01; /* 3FCC71C5 1D8E78AF */
const LG5: f64 = 1.818357216161805012e-01; /* 3FC74664 96CB03DE */
const LG6: f64 = 1.531383769920937332e-01; /* 3FC39A09 D078C69F */
const LG7: f64 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log2(mut x: f64) -> f64 {
let x1p54 = f64::from_bits(0x4350000000000000); // 0x1p54 === 2 ^ 54
let mut ui: u64 = x.to_bits();
let hfsq: f64;
let f: f64;
let s: f64;
let z: f64;
let r: f64;
let mut w: f64;
let t1: f64;
let t2: f64;
let y: f64;
let mut hi: f64;
let lo: f64;
let mut val_hi: f64;
let mut val_lo: f64;
let mut hx: u32;
let mut k: i32;
hx = (ui >> 32) as u32;
k = 0;
if hx < 0x00100000 || (hx >> 31) > 0 {
if ui << 1 == 0 {
return -1. / (x * x); /* log(+-0)=-inf */
}
if (hx >> 31) > 0 {
return (x - x) / 0.0; /* log(-#) = NaN */
}
/* subnormal number, scale x up */
k -= 54;
x *= x1p54;
ui = x.to_bits();
hx = (ui >> 32) as u32;
} else if hx >= 0x7ff00000 {
return x;
} else if hx == 0x3ff00000 && ui << 32 == 0 {
return 0.;
}
/* reduce x into [sqrt(2)/2, sqrt(2)] */
hx += 0x3ff00000 - 0x3fe6a09e;
k += (hx >> 20) as i32 - 0x3ff;
hx = (hx & 0x000fffff) + 0x3fe6a09e;
ui = (hx as u64) << 32 | (ui & 0xffffffff);
x = f64::from_bits(ui);
f = x - 1.0;
hfsq = 0.5 * f * f;
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * (LG4 + w * LG6));
t2 = z * (LG1 + w * (LG3 + w * (LG5 + w * LG7)));
r = t2 + t1;
/* hi+lo = f - hfsq + s*(hfsq+R) ~ log(1+f) */
hi = f - hfsq;
ui = hi.to_bits();
ui &= (-1i64 as u64) << 32;
hi = f64::from_bits(ui);
lo = f - hi - hfsq + s * (hfsq + r);
val_hi = hi * IVLN2HI;
val_lo = (lo + hi) * IVLN2LO + lo * IVLN2HI;
/* spadd(val_hi, val_lo, y), except for not using double_t: */
y = k.into();
w = y + val_hi;
val_lo += (y - w) + val_hi;
val_hi = w;
val_lo + val_hi
}

View File

@ -0,0 +1,87 @@
/* origin: FreeBSD /usr/src/lib/msun/src/e_log2f.c */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* See comments in log2.c.
*/
use core::f32;
const IVLN2HI: f32 = 1.4428710938e+00; /* 0x3fb8b000 */
const IVLN2LO: f32 = -1.7605285393e-04; /* 0xb9389ad4 */
/* |(log(1+s)-log(1-s))/s - Lg(s)| < 2**-34.24 (~[-4.95e-11, 4.97e-11]). */
const LG1: f32 = 0.66666662693; /* 0xaaaaaa.0p-24 */
const LG2: f32 = 0.40000972152; /* 0xccce13.0p-25 */
const LG3: f32 = 0.28498786688; /* 0x91e9ee.0p-25 */
const LG4: f32 = 0.24279078841; /* 0xf89e26.0p-26 */
#[cfg_attr(all(test), no_panic::no_panic)]
pub fn log2f(mut x: f32) -> f32 {
let x1p25f = f32::from_bits(0x4c000000); // 0x1p25f === 2 ^ 25
let mut ui: u32 = x.to_bits();
let hfsq: f32;
let f: f32;
let s: f32;
let z: f32;
let r: f32;
let w: f32;
let t1: f32;
let t2: f32;
let mut hi: f32;
let lo: f32;
let mut ix: u32;
let mut k: i32;
ix = ui;
k = 0;
if ix < 0x00800000 || (ix >> 31) > 0 {
/* x < 2**-126 */
if ix << 1 == 0 {
return -1. / (x * x); /* log(+-0)=-inf */
}
if (ix >> 31) > 0 {
return (x - x) / 0.0; /* log(-#) = NaN */
}
/* subnormal number, scale up x */
k -= 25;
x *= x1p25f;
ui = x.to_bits();
ix = ui;
} else if ix >= 0x7f800000 {
return x;
} else if ix == 0x3f800000 {
return 0.;
}
/* reduce x into [sqrt(2)/2, sqrt(2)] */
ix += 0x3f800000 - 0x3f3504f3;
k += (ix >> 23) as i32 - 0x7f;
ix = (ix & 0x007fffff) + 0x3f3504f3;
ui = ix;
x = f32::from_bits(ui);
f = x - 1.0;
s = f / (2.0 + f);
z = s * s;
w = z * z;
t1 = w * (LG2 + w * LG4);
t2 = z * (LG1 + w * LG3);
r = t2 + t1;
hfsq = 0.5 * f * f;
hi = f - hfsq;
ui = hi.to_bits();
ui &= 0xfffff000;
hi = f32::from_bits(ui);
lo = f - hi - hfsq + s * (hfsq + r);
(lo + hi) * IVLN2LO + lo * IVLN2HI + hi * IVLN2HI + k as f32
}

Some files were not shown because too many files have changed in this diff Show More