Re-vendors wide 0.7 + safe_arch, restores the upstream simd-stable wiring in rapier3d/parry3d manifests and the cfg-simd source, and drops the added 'stripped build does not support SIMD' guards (upstream's simd-vs-enhanced-determinism exclusivity guard kept). Interleaved single-thread retest (min of 4): SIMD buys rapier 1.8-2.2x; box3d vs rapier-simd is now near parity — large_pyramid 1579 vs 1638 ms, joint_grid 957 vs 1008 ms (box3d ahead), many_pyramids 2389 vs 1970 ms (rapier ahead). box3d README grid updated with the honest three-column table; box3d keeps cross-arch determinism + zero deps at that speed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
307 lines
6.6 KiB
Rust
307 lines
6.6 KiB
Rust
use super::*;
|
|
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
#[derive(Default, Clone, Copy, PartialEq, Eq)]
|
|
#[repr(C, align(32))]
|
|
pub struct u8x32 { pub(crate) avx: m256i }
|
|
} else {
|
|
#[derive(Default, Clone, Copy, PartialEq, Eq)]
|
|
#[repr(C, align(32))]
|
|
pub struct u8x32 { pub(crate) a : u8x16, pub(crate) b : u8x16 }
|
|
}
|
|
}
|
|
|
|
int_uint_consts!(u8, 32, u8x32, 256);
|
|
|
|
unsafe impl Zeroable for u8x32 {}
|
|
unsafe impl Pod for u8x32 {}
|
|
|
|
impl Add for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn add(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: add_i8_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.add(rhs.a),
|
|
b : self.b.add(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl Sub for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn sub(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: sub_i8_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.sub(rhs.a),
|
|
b : self.b.sub(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl Add<u8> for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn add(self, rhs: u8) -> Self::Output {
|
|
self.add(Self::splat(rhs))
|
|
}
|
|
}
|
|
|
|
impl Sub<u8> for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn sub(self, rhs: u8) -> Self::Output {
|
|
self.sub(Self::splat(rhs))
|
|
}
|
|
}
|
|
|
|
impl Add<u8x32> for u8 {
|
|
type Output = u8x32;
|
|
#[inline]
|
|
#[must_use]
|
|
fn add(self, rhs: u8x32) -> Self::Output {
|
|
u8x32::splat(self).add(rhs)
|
|
}
|
|
}
|
|
|
|
impl Sub<u8x32> for u8 {
|
|
type Output = u8x32;
|
|
#[inline]
|
|
#[must_use]
|
|
fn sub(self, rhs: u8x32) -> Self::Output {
|
|
u8x32::splat(self).sub(rhs)
|
|
}
|
|
}
|
|
|
|
impl BitAnd for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn bitand(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx : bitand_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.bitand(rhs.a),
|
|
b : self.b.bitand(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl BitOr for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn bitor(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx : bitor_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.bitor(rhs.a),
|
|
b : self.b.bitor(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl BitXor for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn bitxor(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx : bitxor_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.bitxor(rhs.a),
|
|
b : self.b.bitxor(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl CmpEq for u8x32 {
|
|
type Output = Self;
|
|
#[inline]
|
|
#[must_use]
|
|
fn cmp_eq(self, rhs: Self) -> Self::Output {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx : cmp_eq_mask_i8_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.cmp_eq(rhs.a),
|
|
b : self.b.cmp_eq(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl u8x32 {
|
|
#[inline]
|
|
#[must_use]
|
|
pub const fn new(array: [u8; 32]) -> Self {
|
|
unsafe { core::mem::transmute(array) }
|
|
}
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn blend(self, t: Self, f: Self) -> Self {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: blend_varying_i8_m256i(f.avx, t.avx, self.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.blend(t.a, f.a),
|
|
b : self.b.blend(t.b, f.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn max(self, rhs: Self) -> Self {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: max_u8_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.max(rhs.a),
|
|
b : self.b.max(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn min(self, rhs: Self) -> Self {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: min_u8_m256i(self.avx,rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.min(rhs.a),
|
|
b : self.b.min(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn saturating_add(self, rhs: Self) -> Self {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: add_saturating_u8_m256i(self.avx, rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.saturating_add(rhs.a),
|
|
b : self.b.saturating_add(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn saturating_sub(self, rhs: Self) -> Self {
|
|
pick! {
|
|
if #[cfg(target_feature="avx2")] {
|
|
Self { avx: sub_saturating_u8_m256i(self.avx, rhs.avx) }
|
|
} else {
|
|
Self {
|
|
a : self.a.saturating_sub(rhs.a),
|
|
b : self.b.saturating_sub(rhs.b),
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn move_mask(self) -> i32 {
|
|
i8x32::move_mask(cast(self))
|
|
}
|
|
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn any(self) -> bool {
|
|
i8x32::any(cast(self))
|
|
}
|
|
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn all(self) -> bool {
|
|
i8x32::all(cast(self))
|
|
}
|
|
|
|
/// Returns a new vector with lanes selected from the lanes of the first input
|
|
/// vector a specified in the second input vector `rhs`.
|
|
/// The indices i in range `[0, 15]` select the i-th element of `self`. For
|
|
/// indices outside of the range the resulting lane is `0`.
|
|
///
|
|
/// This note that is the equivalent of two parallel swizzle operations on the
|
|
/// two halves of the vector, and the indexes each refer to the
|
|
/// corresponding half.
|
|
#[inline]
|
|
pub fn swizzle_half(self, rhs: i8x32) -> i8x32 {
|
|
cast(i8x32::swizzle_half(cast(self), cast(rhs)))
|
|
}
|
|
|
|
/// Indices in the range `[0, 15]` will select the i-th element of `self`. If
|
|
/// the high bit of any element of `rhs` is set (negative) then the
|
|
/// corresponding output lane is guaranteed to be zero. Otherwise if the
|
|
/// element of `rhs` is within the range `[32, 127]` then the output lane is
|
|
/// either `0` or `self[rhs[i] % 16]` depending on the implementation.
|
|
///
|
|
/// This is the equivalent to two parallel swizzle operations on the two
|
|
/// halves of the vector, and the indexes each refer to their corresponding
|
|
/// half.
|
|
#[inline]
|
|
pub fn swizzle_half_relaxed(self, rhs: u8x32) -> u8x32 {
|
|
cast(i8x32::swizzle_half_relaxed(cast(self), cast(rhs)))
|
|
}
|
|
|
|
#[inline]
|
|
#[must_use]
|
|
pub fn none(self) -> bool {
|
|
!self.any()
|
|
}
|
|
|
|
#[inline]
|
|
pub fn to_array(self) -> [u8; 32] {
|
|
cast(self)
|
|
}
|
|
|
|
#[inline]
|
|
pub fn as_array_ref(&self) -> &[u8; 32] {
|
|
cast_ref(self)
|
|
}
|
|
|
|
#[inline]
|
|
pub fn as_array_mut(&mut self) -> &mut [u8; 32] {
|
|
cast_mut(self)
|
|
}
|
|
}
|