libm

Fixed

• Fix an incorrect result for fminimum and fmaximum with the input (-0, NaN)
• Fix a typo in libm::Libm::roundeven
• Fix the expm1f overflow threshold
• Change CmpResult to use a pointer-sized return type
• Compare against CARGO_CFG_TARGET_FAMILY in a multi-valued fashion
• Implement exp and its variants for i586 with inline assembly
• Implement floor and ceil in assembly on i586

Other

• Significantly optimize fmod worst case performance (#1002)

Fixed

• Fix an incorrect result for fminimum and fmaximum with the input (-0, NaN)
• Fix a typo in libm::Libm::roundeven
• Fix the expm1f overflow threshold
• Change CmpResult to use a pointer-sized return type
• Compare against CARGO_CFG_TARGET_FAMILY in a multi-valued fashion
• Implement exp and its variants for i586 with inline assembly
• Implement floor and ceil in assembly on i586

Other

• Significantly optimize fmod worst case performance (#1002)

Other

• Require target_has_atomic = "ptr" for runtime feature detection

Other

• Require target_has_atomic = "ptr" for runtime feature detection

Other

• Use runtime feature detection for fma routines on x86

Other

• Use runtime feature detection for fma routines on x86

Fixed

• Switch back to workspace resolver v2 to unbreak builds without the 2024 edition

Fixed

• Switch back to workspace resolver v2 to unbreak builds without the 2024 edition

• Mark generic functions #[inline]
• Combine the source files for fmod
• Ensure all public functions are marked no_panic
• Add assembly version of simple operations on aarch64
• Add roundeven{,f,f16,f128}
• Add fminimum, fmaximum, fminimum_num, and fmaximum_num
• Eliminate the use of force_eval! in ceil, floor, and trunc
• Port the CORE-MATH version of cbrt
• Add fmaf128
• fma: Ensure zero has the correct sign
• Add scalbnf16, scalbnf128, ldexpf16, and ldexpf128
• Specify license as just MIT
• Add fmodf128
• Add fmodf16 using the generic implementation
• Add fminf16, fmaxf16, fminf128, and fmaxf128
• Add roundf16 and roundf128
• Add rintf16 and rintf128
• Add floorf16 and floorf128
• Add ceilf16 and ceilf128
• Add sqrtf16 and sqrtf128
• Simplify and optimize fdim (#442)
• Add fdimf16 and fdimf128
• Add truncf16 and truncf128
• Add fabsf16, fabsf128, copysignf16, and copysignf128
• Move some numeric trait logic to default implementations
• Add some more basic docstrings (#352)
• Add support for loongarch64-unknown-linux-gnu
• Add an "arch" Cargo feature that is on by default
• Rename the special_case module to precision and move default ULP
• Move the existing "unstable" feature to "unstable-intrinsics"

There are a number of things that changed internally, see the git log for a full
list of changes.

• Mark generic functions #[inline]
• Combine the source files for fmod
• Ensure all public functions are marked no_panic
• Add assembly version of simple operations on aarch64
• Add roundeven{,f,f16,f128}
• Add fminimum, fmaximum, fminimum_num, and fmaximum_num
• Eliminate the use of force_eval! in ceil, floor, and trunc
• Port the CORE-MATH version of cbrt
• Add fmaf128
• fma: Ensure zero has the correct sign
• Add scalbnf16, scalbnf128, ldexpf16, and ldexpf128
• Specify license as just MIT
• Add fmodf128
• Add fmodf16 using the generic implementation
• Add fminf16, fmaxf16, fminf128, and fmaxf128
• Add roundf16 and roundf128
• Add rintf16 and rintf128
• Add floorf16 and floorf128
• Add ceilf16 and ceilf128
• Add sqrtf16 and sqrtf128
• Simplify and optimize fdim (#442)
• Add fdimf16 and fdimf128
• Add truncf16 and truncf128
• Add fabsf16, fabsf128, copysignf16, and copysignf128
• Move some numeric trait logic to default implementations
• Add some more basic docstrings (#352)
• Add support for loongarch64-unknown-linux-gnu
• Add an "arch" Cargo feature that is on by default
• Rename the special_case module to precision and move default ULP
• Move the existing "unstable" feature to "unstable-intrinsics"

There are a number of things that changed internally, see the git log for a full list of changes.

Fixed

• fix type of constants in ported sincosf (#331)

Other

• Disable a unit test that is failing on i586
• Add a procedural macro for expanding all function signatures
• Introduce musl-math-sys for bindings to musl math symbols
• Add basic docstrings to some functions (#337)

Other

• Set the MSRV to 1.63 and test this in CI

Fixed

• Update exponent calculations in nextafter to match musl

Changed

• Update licensing to MIT AND (MIT OR Apache-2.0), as this is derivative from MIT-licensed musl.
• Set edition to 2021 for all crates
• Upgrade all dependencies

Other

• Don't deny warnings in lib.rs
• Rename the musl-bitwise-tests feature to test-musl-serialized
• Rename the musl-reference-tests feature to musl-bitwise-tests
• Move musl-reference-tests to a new libm-test crate
• Add a force-soft-floats feature to prevent using any intrinsics or arch-specific code
• Deny warnings in CI
• Fix clippy::deprecated_cfg_attr on compiler_builtins
• Corrected English typos
• Remove unneeded extern core in tgamma
• Allow internal_features lint when building with "unstable"

Fixed

• sincosf

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

Fixed

• Restored compatibility with Rust 1.31.0

Added

• minf
• fmin
• fmaxf
• fmax

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

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

• Initial release