From c31a097d7533a1df9435119e40f649b88a0d387e Mon Sep 17 00:00:00 2001 From: Henry Jiang Date: Fri, 22 Nov 2024 17:47:15 -0500 Subject: [PATCH] aix: create shim for lgammaf_r --- library/std/src/sys/cmath.rs | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/library/std/src/sys/cmath.rs b/library/std/src/sys/cmath.rs index 2997e908fa1b2..829042381a24c 100644 --- a/library/std/src/sys/cmath.rs +++ b/library/std/src/sys/cmath.rs @@ -26,6 +26,7 @@ extern "C" { pub fn tgamma(n: f64) -> f64; pub fn tgammaf(n: f32) -> f32; pub fn lgamma_r(n: f64, s: &mut i32) -> f64; + #[cfg(not(target_os = "aix"))] pub fn lgammaf_r(n: f32, s: &mut i32) -> f32; pub fn acosf128(n: f128) -> f128; @@ -56,13 +57,20 @@ extern "C" { }} } +// On AIX, we don't have lgammaf_r only the f64 version, so we can +// use the f64 version lgamma_r +#[cfg(target_os = "aix")] +pub unsafe fn lgammaf_r(n: f32, s: &mut i32) -> f32 { + lgamma_r(n.into(), s) as f32 +} + // On 32-bit x86 MSVC these functions aren't defined, so we just define shims // which promote everything to f64, perform the calculation, and then demote // back to f32. While not precisely correct should be "correct enough" for now. cfg_if::cfg_if! { if #[cfg(all(target_os = "windows", target_env = "msvc", target_arch = "x86"))] { #[inline] - pub unsafe fn acosf(n: f32) -> f32 { + pub unsafe fn acosf(n: f32) -> f32 { f64::acos(n as f64) as f32 }