Skip to content

cleanup FlushSubnormals implementation #538

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
folkertdev wants to merge 1 commit into
base: master
Choose a base branch
from
Open

cleanup FlushSubnormals implementation #538

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
106 changes: 35 additions & 71 deletions crates/test_helpers/src/subnormals.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,81 +7,45 @@ pub trait FlushSubnormals: Sized {
impl<T> FlushSubnormals for *const T {}
impl<T> FlushSubnormals for *mut T {}

macro_rules! impl_float {
{ $($ty:ty),* } => {
$(
impl FlushSubnormals for $ty {
fn flush(self) -> Self {
let is_f32 = size_of::<Self>() == 4;
let ppc_flush = is_f32 && cfg!(all(
any(target_arch = "powerpc", all(target_arch = "powerpc64", target_endian = "big")),
target_feature = "altivec",
not(target_feature = "vsx"),
));
let arm_flush = is_f32 && cfg!(all(target_arch = "arm", target_feature = "neon"));
let flush = ppc_flush || arm_flush;
if flush && self.is_subnormal() {
<$ty>::copysign(0., self)
} else {
self
}
}
impl FlushSubnormals for i8 {}
impl FlushSubnormals for i16 {}
impl FlushSubnormals for i32 {}
impl FlushSubnormals for i64 {}
impl FlushSubnormals for isize {}

impl FlushSubnormals for u8 {}
impl FlushSubnormals for u16 {}
impl FlushSubnormals for u32 {}
impl FlushSubnormals for u64 {}
impl FlushSubnormals for usize {}

impl FlushSubnormals for f16 {}
impl FlushSubnormals for f64 {}

impl FlushSubnormals for f32 {
fn flush(self) -> Self {
// Correct for non-IEEE behavior for f32 on arm and powerpc.
let ppc_flush = cfg!(all(
any(
target_arch = "powerpc",
all(target_arch = "powerpc64", target_endian = "big")
),
target_feature = "altivec",
not(target_feature = "vsx"),
));
let arm_flush = cfg!(all(target_arch = "arm", target_feature = "neon"));
let flush = ppc_flush || arm_flush;

if flush && self.is_subnormal() {
f32::copysign(0.0, self)
} else {
self
}
)*
}
}

macro_rules! impl_else {
{ $($ty:ty),* } => {
$(
impl FlushSubnormals for $ty {}
)*
}
}

impl_float! { f16, f32, f64 }
impl_else! { i8, i16, i32, i64, isize, u8, u16, u32, u64, usize }

/// AltiVec should flush subnormal inputs to zero, but QEMU seems to only flush outputs.
/// https://gitlab.com/qemu-project/qemu/-/issues/1779
#[cfg(all(
any(target_arch = "powerpc", target_arch = "powerpc64"),
target_feature = "altivec"
))]
fn in_buggy_qemu() -> bool {
use std::sync::OnceLock;
static BUGGY: OnceLock<bool> = OnceLock::new();

fn add(x: f32, y: f32) -> f32 {
#[cfg(target_arch = "powerpc")]
use core::arch::powerpc::*;
#[cfg(target_arch = "powerpc64")]
use core::arch::powerpc64::*;

let array: [f32; 4] =
unsafe { core::mem::transmute(vec_add(vec_splats(x), vec_splats(y))) };
array[0]
}

*BUGGY.get_or_init(|| add(-1.0857398e-38, 0.).is_sign_negative())
}

#[cfg(all(
any(target_arch = "powerpc", target_arch = "powerpc64"),
target_feature = "altivec"
))]
pub fn flush_in<T: FlushSubnormals>(x: T) -> T {
if in_buggy_qemu() {
x
} else {
x.flush()
}
}

#[cfg(not(all(
any(target_arch = "powerpc", target_arch = "powerpc64"),
target_feature = "altivec"
)))]
/// NOTE: altivec had a subnormal flushing bug in older QEMU versions.
/// <https://gitlab.com/qemu-project/qemu/-/issues/1779>
pub fn flush_in<T: FlushSubnormals>(x: T) -> T {
x.flush()
}
Expand Down
Loading