From 97179e46a8fbbda0053a3fce69ed3fd5d488acdc Mon Sep 17 00:00:00 2001
From: beetrees <b@beetr.ee>
Date: Fri, 23 Aug 2024 12:39:31 +0100
Subject: [PATCH] Fix ABI for `f16` builtins on Intel Apple targets
---
src/float/extend.rs | 2 +
src/float/trunc.rs | 3 ++
src/macros.rs | 100 ++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 105 insertions(+)
diff --git a/src/float/extend.rs b/src/float/extend.rs
index 2f392255f..9fabcde25 100644
--- a/src/float/extend.rs
+++ b/src/float/extend.rs
@@ -86,6 +86,7 @@ intrinsics! {
intrinsics! {
#[avr_skip]
#[aapcs_on_arm]
+ #[apple_f16_arg_abi]
#[arm_aeabi_alias = __aeabi_h2f]
#[cfg(f16_enabled)]
pub extern "C" fn __extendhfsf2(a: f16) -> f32 {
@@ -94,6 +95,7 @@ intrinsics! {
#[avr_skip]
#[aapcs_on_arm]
+ #[apple_f16_arg_abi]
#[cfg(f16_enabled)]
pub extern "C" fn __gnu_h2f_ieee(a: f16) -> f32 {
extend(a)
diff --git a/src/float/trunc.rs b/src/float/trunc.rs
index c54ff7805..5c17cd96a 100644
--- a/src/float/trunc.rs
+++ b/src/float/trunc.rs
@@ -134,6 +134,7 @@ intrinsics! {
intrinsics! {
#[avr_skip]
#[aapcs_on_arm]
+ #[apple_f16_ret_abi]
#[arm_aeabi_alias = __aeabi_f2h]
#[cfg(f16_enabled)]
pub extern "C" fn __truncsfhf2(a: f32) -> f16 {
@@ -142,6 +143,7 @@ intrinsics! {
#[avr_skip]
#[aapcs_on_arm]
+ #[apple_f16_ret_abi]
#[cfg(f16_enabled)]
pub extern "C" fn __gnu_f2h_ieee(a: f32) -> f16 {
trunc(a)
@@ -149,6 +151,7 @@ intrinsics! {
#[avr_skip]
#[aapcs_on_arm]
+ #[apple_f16_ret_abi]
#[arm_aeabi_alias = __aeabi_d2h]
#[cfg(f16_enabled)]
pub extern "C" fn __truncdfhf2(a: f64) -> f16 {
diff --git a/src/macros.rs b/src/macros.rs
index f7c6d0bfe..f51e49e98 100644
--- a/src/macros.rs
+++ b/src/macros.rs
@@ -276,6 +276,106 @@ macro_rules! intrinsics {
intrinsics!($($rest)*);
);
+ // `arm_aeabi_alias` would conflict with `f16_apple_{arg,ret}_abi` not handled here. Avoid macro ambiguity by combining in a
+ // single `#[]`.
+ (
+ #[apple_f16_arg_abi]
+ #[arm_aeabi_alias = $alias:ident]
+ $($t:tt)*
+ ) => {
+ intrinsics! {
+ #[apple_f16_arg_abi, arm_aeabi_alias = $alias]
+ $($t)*
+ }
+ };
+ (
+ #[apple_f16_ret_abi]
+ #[arm_aeabi_alias = $alias:ident]
+ $($t:tt)*
+ ) => {
+ intrinsics! {
+ #[apple_f16_ret_abi, arm_aeabi_alias = $alias]
+ $($t)*
+ }
+ };
+
+ // On x86 (32-bit and 64-bit) Apple platforms, `f16` is passed and returned like a `u16` unless
+ // the builtin involves `f128`.
+ (
+ // `arm_aeabi_alias` would conflict if not handled here. Avoid macro ambiguity by combining
+ // in a single `#[]`.
+ #[apple_f16_arg_abi $(, arm_aeabi_alias = $alias:ident)?]
+ $(#[$($attr:tt)*])*
+ pub extern $abi:tt fn $name:ident( $($argname:ident: $ty:ty),* ) $(-> $ret:ty)? {
+ $($body:tt)*
+ }
+
+ $($rest:tt)*
+ ) => (
+ #[cfg(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64")))]
+ $(#[$($attr)*])*
+ pub extern $abi fn $name( $($argname: $ty),* ) $(-> $ret)? {
+ $($body)*
+ }
+
+ #[cfg(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64"), not(feature = "mangled-names")))]
+ mod $name {
+ #[no_mangle]
+ #[cfg_attr(not(all(windows, target_env = "gnu")), linkage = "weak")]
+ $(#[$($attr)*])*
+ extern $abi fn $name( $($argname: u16),* ) $(-> $ret)? {
+ super::$name($(f16::from_bits($argname)),*)
+ }
+ }
+
+ #[cfg(not(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64"))))]
+ intrinsics! {
+ $(#[arm_aeabi_alias = $alias])?
+ $(#[$($attr)*])*
+ pub extern $abi fn $name( $($argname: $ty),* ) $(-> $ret)? {
+ $($body)*
+ }
+ }
+
+ intrinsics!($($rest)*);
+ );
+ (
+ #[apple_f16_ret_abi $(, arm_aeabi_alias = $alias:ident)?]
+ $(#[$($attr:tt)*])*
+ pub extern $abi:tt fn $name:ident( $($argname:ident: $ty:ty),* ) $(-> $ret:ty)? {
+ $($body:tt)*
+ }
+
+ $($rest:tt)*
+ ) => (
+ #[cfg(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64")))]
+ $(#[$($attr)*])*
+ pub extern $abi fn $name( $($argname: $ty),* ) $(-> $ret)? {
+ $($body)*
+ }
+
+ #[cfg(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64"), not(feature = "mangled-names")))]
+ mod $name {
+ #[no_mangle]
+ #[cfg_attr(not(all(windows, target_env = "gnu")), linkage = "weak")]
+ $(#[$($attr)*])*
+ extern $abi fn $name( $($argname: $ty),* ) -> u16 {
+ super::$name($($argname),*).to_bits()
+ }
+ }
+
+ #[cfg(not(all(target_vendor = "apple", any(target_arch = "x86", target_arch = "x86_64"))))]
+ intrinsics! {
+ $(#[arm_aeabi_alias = $alias])?
+ $(#[$($attr)*])*
+ pub extern $abi fn $name( $($argname: $ty),* ) $(-> $ret)? {
+ $($body)*
+ }
+ }
+
+ intrinsics!($($rest)*);
+ );
+
// A bunch of intrinsics on ARM are aliased in the standard compiler-rt
// build under `__aeabi_*` aliases, and LLVM will call these instead of the
// original function. The aliasing here is used to generate these symbols in