use num_traits::MulAdd;
use std::ops::{Add, Mul, Neg};
#[cfg(any(
all(
any(target_arch = "x86", target_arch = "x86_64"),
target_feature = "fma"
),
all(target_arch = "aarch64", target_feature = "neon")
))]
#[inline(always)]
pub(crate) fn mlaf<T: Copy + Mul<T, Output = T> + Add<T, Output = T> + MulAdd<T, Output = T>>(
acc: T,
a: T,
b: T,
) -> T {
MulAdd::mul_add(a, b, acc)
}
#[inline(always)]
#[cfg(not(any(
all(
any(target_arch = "x86", target_arch = "x86_64"),
target_feature = "fma"
),
all(target_arch = "aarch64", target_feature = "neon")
)))]
pub(crate) fn mlaf<T: Copy + Mul<T, Output = T> + Add<T, Output = T> + MulAdd<T, Output = T>>(
acc: T,
a: T,
b: T,
) -> T {
acc + a * b
}
#[inline(always)]
pub(crate) fn neg_mlaf<
T: Copy + Mul<T, Output = T> + Add<T, Output = T> + MulAdd<T, Output = T> + Neg<Output = T>,
>(
acc: T,
a: T,
b: T,
) -> T {
mlaf(acc, a, -b)
}
#[inline(always)]
pub(crate) fn fmla<T: Copy + Mul<T, Output = T> + Add<T, Output = T> + MulAdd<T, Output = T>>(
a: T,
b: T,
acc: T,
) -> T {
mlaf(acc, a, b)
}