Implement core::ops (#10)

* Add vector-vector arithmetic ops
* Add operators and integer conversions for masks
* Add unary traits
* Implement Index and IndexMut
* Implement by-ref ops for masks
* Document intrinsics
* Implement format traits for masks
* Add floating point ops tests
* Add integer tests
* Add mask tests
This commit is contained in:
Caleb Zulawski
2020-10-01 22:50:15 -04:00
committed by GitHub
parent fa6bb81e44
commit 43dabd1aea
33 changed files with 2233 additions and 1 deletions

View File

@@ -0,0 +1,39 @@
//! This module contains the LLVM intrinsics bindings that provide the functionality for this
//! crate.
//!
//! The LLVM assembly language is documented here: https://llvm.org/docs/LangRef.html
/// These intrinsics aren't linked directly from LLVM and are mostly undocumented, however they are
/// simply lowered to the matching LLVM instructions by the compiler. The associated instruction
/// is documented alongside each intrinsic.
extern "platform-intrinsic" {
/// add/fadd
pub(crate) fn simd_add<T>(x: T, y: T) -> T;
/// sub/fsub
pub(crate) fn simd_sub<T>(x: T, y: T) -> T;
/// mul/fmul
pub(crate) fn simd_mul<T>(x: T, y: T) -> T;
/// udiv/sdiv/fdiv
pub(crate) fn simd_div<T>(x: T, y: T) -> T;
/// urem/srem/frem
pub(crate) fn simd_rem<T>(x: T, y: T) -> T;
/// shl
pub(crate) fn simd_shl<T>(x: T, y: T) -> T;
/// lshr/ashr
pub(crate) fn simd_shr<T>(x: T, y: T) -> T;
/// and
pub(crate) fn simd_and<T>(x: T, y: T) -> T;
/// or
pub(crate) fn simd_or<T>(x: T, y: T) -> T;
/// xor
pub(crate) fn simd_xor<T>(x: T, y: T) -> T;
}

View File

@@ -1,5 +1,5 @@
#![no_std]
#![feature(repr_simd)]
#![feature(repr_simd, platform_intrinsics)]
#![warn(missing_docs)]
//! Portable SIMD module.
@@ -7,6 +7,8 @@
mod macros;
mod fmt;
mod intrinsics;
mod ops;
mod masks;
pub use masks::*;

View File

@@ -1,3 +1,13 @@
/// The error type returned when converting an integer to a mask fails.
#[derive(Debug, Copy, Clone, PartialEq, Eq)]
pub struct TryFromMaskError(());
impl core::fmt::Display for TryFromMaskError {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
write!(f, "mask must have all bits set or unset")
}
}
macro_rules! define_mask {
{ $(#[$attr:meta])* struct $name:ident($type:ty); } => {
$(#[$attr])*
@@ -34,11 +44,52 @@ macro_rules! define_mask {
}
}
impl core::convert::TryFrom<$type> for $name {
type Error = TryFromMaskError;
fn try_from(value: $type) -> Result<Self, Self::Error> {
if value == 0 || !value == 0 {
Ok(Self(value))
} else {
Err(TryFromMaskError(()))
}
}
}
impl core::convert::From<$name> for $type {
fn from(value: $name) -> Self {
value.0
}
}
impl core::fmt::Debug for $name {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
self.test().fmt(f)
}
}
impl core::fmt::Binary for $name {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
<$type as core::fmt::Binary>::fmt(&self.0, f)
}
}
impl core::fmt::Octal for $name {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
<$type as core::fmt::Octal>::fmt(&self.0, f)
}
}
impl core::fmt::LowerHex for $name {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
<$type as core::fmt::LowerHex>::fmt(&self.0, f)
}
}
impl core::fmt::UpperHex for $name {
fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
<$type as core::fmt::UpperHex>::fmt(&self.0, f)
}
}
}
}

628
crates/core_simd/src/ops.rs Normal file
View File

@@ -0,0 +1,628 @@
/// Checks if the right-hand side argument of a left- or right-shift would cause overflow.
fn invalid_shift_rhs<T>(rhs: T) -> bool
where
T: Default + PartialOrd + core::convert::TryFrom<usize>,
<T as core::convert::TryFrom<usize>>::Error: core::fmt::Debug,
{
let bits_in_type = T::try_from(8 * core::mem::size_of::<T>()).unwrap();
rhs < T::default() || rhs >= bits_in_type
}
/// Automatically implements operators over references in addition to the provided operator.
macro_rules! impl_ref_ops {
// binary op
{
impl core::ops::$trait:ident<$rhs:ty> for $type:ty {
type Output = $output:ty;
$(#[$attrs:meta])*
fn $fn:ident($self_tok:ident, $rhs_arg:ident: $rhs_arg_ty:ty) -> Self::Output $body:tt
}
} => {
impl core::ops::$trait<$rhs> for $type {
type Output = $output;
$(#[$attrs])*
fn $fn($self_tok, $rhs_arg: $rhs_arg_ty) -> Self::Output $body
}
impl core::ops::$trait<&'_ $rhs> for $type {
type Output = <$type as core::ops::$trait<$rhs>>::Output;
$(#[$attrs])*
fn $fn($self_tok, $rhs_arg: &$rhs) -> Self::Output {
core::ops::$trait::$fn($self_tok, *$rhs_arg)
}
}
impl core::ops::$trait<$rhs> for &'_ $type {
type Output = <$type as core::ops::$trait<$rhs>>::Output;
$(#[$attrs])*
fn $fn($self_tok, $rhs_arg: $rhs) -> Self::Output {
core::ops::$trait::$fn(*$self_tok, $rhs_arg)
}
}
impl core::ops::$trait<&'_ $rhs> for &'_ $type {
type Output = <$type as core::ops::$trait<$rhs>>::Output;
$(#[$attrs])*
fn $fn($self_tok, $rhs_arg: &$rhs) -> Self::Output {
core::ops::$trait::$fn(*$self_tok, *$rhs_arg)
}
}
};
// binary assignment op
{
impl core::ops::$trait:ident<$rhs:ty> for $type:ty {
$(#[$attrs:meta])*
fn $fn:ident(&mut $self_tok:ident, $rhs_arg:ident: $rhs_arg_ty:ty) $body:tt
}
} => {
impl core::ops::$trait<$rhs> for $type {
$(#[$attrs])*
fn $fn(&mut $self_tok, $rhs_arg: $rhs_arg_ty) $body
}
impl core::ops::$trait<&'_ $rhs> for $type {
$(#[$attrs])*
fn $fn(&mut $self_tok, $rhs_arg: &$rhs_arg_ty) {
core::ops::$trait::$fn($self_tok, *$rhs_arg)
}
}
};
// unary op
{
impl core::ops::$trait:ident for $type:ty {
type Output = $output:ty;
fn $fn:ident($self_tok:ident) -> Self::Output $body:tt
}
} => {
impl core::ops::$trait for $type {
type Output = $output;
fn $fn($self_tok) -> Self::Output $body
}
impl core::ops::$trait for &'_ $type {
type Output = <$type as core::ops::$trait>::Output;
fn $fn($self_tok) -> Self::Output {
core::ops::$trait::$fn(*$self_tok)
}
}
}
}
/// Implements op traits for masks
macro_rules! impl_mask_ops {
{ $($mask:ty),* } => {
$(
impl_ref_ops! {
impl core::ops::BitAnd<$mask> for $mask {
type Output = Self;
fn bitand(self, rhs: Self) -> Self::Output {
Self(self.0 & rhs.0)
}
}
}
impl_ref_ops! {
impl core::ops::BitAndAssign<$mask> for $mask {
fn bitand_assign(&mut self, rhs: Self) {
*self = *self & rhs;
}
}
}
impl_ref_ops! {
impl core::ops::BitOr<$mask> for $mask {
type Output = Self;
fn bitor(self, rhs: Self) -> Self::Output {
Self(self.0 | rhs.0)
}
}
}
impl_ref_ops! {
impl core::ops::BitOrAssign<$mask> for $mask {
fn bitor_assign(&mut self, rhs: Self) {
*self = *self | rhs;
}
}
}
impl_ref_ops! {
impl core::ops::BitXor<$mask> for $mask {
type Output = Self;
fn bitxor(self, rhs: Self) -> Self::Output {
Self(self.0 ^ rhs.0)
}
}
}
impl_ref_ops! {
impl core::ops::BitXorAssign<$mask> for $mask {
fn bitxor_assign(&mut self, rhs: Self) {
*self = *self ^ rhs;
}
}
}
impl_ref_ops! {
impl core::ops::Not for $mask {
type Output = Self;
fn not(self) -> Self::Output {
Self(!self.0)
}
}
}
)*
}
}
impl_mask_ops! { crate::mask8, crate::mask16, crate::mask32, crate::mask64, crate::mask128, crate::masksize }
/// Automatically implements operators over vectors and scalars for a particular vector.
macro_rules! impl_op {
{ impl Add for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Add::add, AddAssign::add_assign, simd_add }
};
{ impl Sub for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Sub::sub, SubAssign::sub_assign, simd_sub }
};
{ impl Mul for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Mul::mul, MulAssign::mul_assign, simd_mul }
};
{ impl Div for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Div::div, DivAssign::div_assign, simd_div }
};
{ impl Rem for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Rem::rem, RemAssign::rem_assign, simd_rem }
};
{ impl Shl for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Shl::shl, ShlAssign::shl_assign, simd_shl }
};
{ impl Shr for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, Shr::shr, ShrAssign::shr_assign, simd_shr }
};
{ impl BitAnd for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, BitAnd::bitand, BitAndAssign::bitand_assign, simd_and }
};
{ impl BitOr for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, BitOr::bitor, BitOrAssign::bitor_assign, simd_or }
};
{ impl BitXor for $type:ty, $scalar:ty } => {
impl_op! { @binary $type, $scalar, BitXor::bitxor, BitXorAssign::bitxor_assign, simd_xor }
};
{ impl Not for $type:ty, $scalar:ty } => {
impl_ref_ops! {
impl core::ops::Not for $type {
type Output = Self;
fn not(self) -> Self::Output {
self ^ <$type>::splat(!<$scalar>::default())
}
}
}
};
{ impl Neg for $type:ty, $scalar:ty } => {
impl_ref_ops! {
impl core::ops::Neg for $type {
type Output = Self;
fn neg(self) -> Self::Output {
<$type>::splat(-<$scalar>::default()) - self
}
}
}
};
{ impl Index for $type:ty, $scalar:ty } => {
impl<I> core::ops::Index<I> for $type
where
I: core::slice::SliceIndex<[$scalar]>,
{
type Output = I::Output;
fn index(&self, index: I) -> &Self::Output {
let slice: &[_] = self.as_ref();
&slice[index]
}
}
impl<I> core::ops::IndexMut<I> for $type
where
I: core::slice::SliceIndex<[$scalar]>,
{
fn index_mut(&mut self, index: I) -> &mut Self::Output {
let slice: &mut [_] = self.as_mut();
&mut slice[index]
}
}
};
// generic binary op with assignment when output is `Self`
{ @binary $type:ty, $scalar:ty, $trait:ident :: $trait_fn:ident, $assign_trait:ident :: $assign_trait_fn:ident, $intrinsic:ident } => {
impl_ref_ops! {
impl core::ops::$trait<$type> for $type {
type Output = $type;
#[inline]
fn $trait_fn(self, rhs: $type) -> Self::Output {
unsafe {
crate::intrinsics::$intrinsic(self, rhs)
}
}
}
}
impl_ref_ops! {
impl core::ops::$trait<$scalar> for $type {
type Output = $type;
#[inline]
fn $trait_fn(self, rhs: $scalar) -> Self::Output {
core::ops::$trait::$trait_fn(self, <$type>::splat(rhs))
}
}
}
impl_ref_ops! {
impl core::ops::$trait<$type> for $scalar {
type Output = $type;
#[inline]
fn $trait_fn(self, rhs: $type) -> Self::Output {
core::ops::$trait::$trait_fn(<$type>::splat(self), rhs)
}
}
}
impl_ref_ops! {
impl core::ops::$assign_trait<$type> for $type {
#[inline]
fn $assign_trait_fn(&mut self, rhs: $type) {
unsafe {
*self = crate::intrinsics::$intrinsic(*self, rhs);
}
}
}
}
impl_ref_ops! {
impl core::ops::$assign_trait<$scalar> for $type {
#[inline]
fn $assign_trait_fn(&mut self, rhs: $scalar) {
core::ops::$assign_trait::$assign_trait_fn(self, <$type>::splat(rhs));
}
}
}
};
}
/// Implements floating-point operators for the provided types.
macro_rules! impl_float_ops {
{ $($scalar:ty => $($vector:ty),*;)* } => {
$( // scalar
$( // vector
impl_op! { impl Add for $vector, $scalar }
impl_op! { impl Sub for $vector, $scalar }
impl_op! { impl Mul for $vector, $scalar }
impl_op! { impl Div for $vector, $scalar }
impl_op! { impl Rem for $vector, $scalar }
impl_op! { impl Neg for $vector, $scalar }
impl_op! { impl Index for $vector, $scalar }
)*
)*
};
}
/// Implements mask operators for the provided types.
macro_rules! impl_mask_ops {
{ $($scalar:ty => $($vector:ty),*;)* } => {
$( // scalar
$( // vector
impl_op! { impl BitAnd for $vector, $scalar }
impl_op! { impl BitOr for $vector, $scalar }
impl_op! { impl BitXor for $vector, $scalar }
impl_op! { impl Not for $vector, $scalar }
impl_op! { impl Index for $vector, $scalar }
)*
)*
};
}
/// Implements unsigned integer operators for the provided types.
macro_rules! impl_unsigned_int_ops {
{ $($scalar:ty => $($vector:ty),*;)* } => {
$( // scalar
$( // vector
impl_op! { impl Add for $vector, $scalar }
impl_op! { impl Sub for $vector, $scalar }
impl_op! { impl Mul for $vector, $scalar }
impl_op! { impl BitAnd for $vector, $scalar }
impl_op! { impl BitOr for $vector, $scalar }
impl_op! { impl BitXor for $vector, $scalar }
impl_op! { impl Not for $vector, $scalar }
impl_op! { impl Index for $vector, $scalar }
// Integers panic on divide by 0
impl_ref_ops! {
impl core::ops::Div<$vector> for $vector {
type Output = Self;
#[inline]
fn div(self, rhs: $vector) -> Self::Output {
// TODO there is probably a better way of doing this
if AsRef::<[$scalar]>::as_ref(&rhs)
.iter()
.any(|x| *x == 0)
{
panic!("attempt to divide by zero");
}
unsafe { crate::intrinsics::simd_div(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Div<$scalar> for $vector {
type Output = $vector;
#[inline]
fn div(self, rhs: $scalar) -> Self::Output {
if rhs == 0 {
panic!("attempt to divide by zero");
}
let rhs = Self::splat(rhs);
unsafe { crate::intrinsics::simd_div(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Div<$vector> for $scalar {
type Output = $vector;
#[inline]
fn div(self, rhs: $vector) -> Self::Output {
<$vector>::splat(self) / rhs
}
}
}
impl_ref_ops! {
impl core::ops::DivAssign<$vector> for $vector {
#[inline]
fn div_assign(&mut self, rhs: Self) {
*self = *self / rhs;
}
}
}
impl_ref_ops! {
impl core::ops::DivAssign<$scalar> for $vector {
#[inline]
fn div_assign(&mut self, rhs: $scalar) {
*self = *self / rhs;
}
}
}
// remainder panics on zero divisor
impl_ref_ops! {
impl core::ops::Rem<$vector> for $vector {
type Output = Self;
#[inline]
fn rem(self, rhs: $vector) -> Self::Output {
// TODO there is probably a better way of doing this
if AsRef::<[$scalar]>::as_ref(&rhs)
.iter()
.any(|x| *x == 0)
{
panic!("attempt to calculate the remainder with a divisor of zero");
}
unsafe { crate::intrinsics::simd_rem(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Rem<$scalar> for $vector {
type Output = $vector;
#[inline]
fn rem(self, rhs: $scalar) -> Self::Output {
if rhs == 0 {
panic!("attempt to calculate the remainder with a divisor of zero");
}
let rhs = Self::splat(rhs);
unsafe { crate::intrinsics::simd_rem(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Rem<$vector> for $scalar {
type Output = $vector;
#[inline]
fn rem(self, rhs: $vector) -> Self::Output {
<$vector>::splat(self) % rhs
}
}
}
impl_ref_ops! {
impl core::ops::RemAssign<$vector> for $vector {
#[inline]
fn rem_assign(&mut self, rhs: Self) {
*self = *self % rhs;
}
}
}
impl_ref_ops! {
impl core::ops::RemAssign<$scalar> for $vector {
#[inline]
fn rem_assign(&mut self, rhs: $scalar) {
*self = *self % rhs;
}
}
}
// shifts panic on overflow
impl_ref_ops! {
impl core::ops::Shl<$vector> for $vector {
type Output = Self;
#[inline]
fn shl(self, rhs: $vector) -> Self::Output {
// TODO there is probably a better way of doing this
if AsRef::<[$scalar]>::as_ref(&rhs)
.iter()
.copied()
.any(invalid_shift_rhs)
{
panic!("attempt to shift left with overflow");
}
unsafe { crate::intrinsics::simd_shl(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Shl<$scalar> for $vector {
type Output = $vector;
#[inline]
fn shl(self, rhs: $scalar) -> Self::Output {
if invalid_shift_rhs(rhs) {
panic!("attempt to shift left with overflow");
}
let rhs = Self::splat(rhs);
unsafe { crate::intrinsics::simd_shl(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::ShlAssign<$vector> for $vector {
#[inline]
fn shl_assign(&mut self, rhs: Self) {
*self = *self << rhs;
}
}
}
impl_ref_ops! {
impl core::ops::ShlAssign<$scalar> for $vector {
#[inline]
fn shl_assign(&mut self, rhs: $scalar) {
*self = *self << rhs;
}
}
}
impl_ref_ops! {
impl core::ops::Shr<$vector> for $vector {
type Output = Self;
#[inline]
fn shr(self, rhs: $vector) -> Self::Output {
// TODO there is probably a better way of doing this
if AsRef::<[$scalar]>::as_ref(&rhs)
.iter()
.copied()
.any(invalid_shift_rhs)
{
panic!("attempt to shift with overflow");
}
unsafe { crate::intrinsics::simd_shr(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::Shr<$scalar> for $vector {
type Output = $vector;
#[inline]
fn shr(self, rhs: $scalar) -> Self::Output {
if invalid_shift_rhs(rhs) {
panic!("attempt to shift with overflow");
}
let rhs = Self::splat(rhs);
unsafe { crate::intrinsics::simd_shr(self, rhs) }
}
}
}
impl_ref_ops! {
impl core::ops::ShrAssign<$vector> for $vector {
#[inline]
fn shr_assign(&mut self, rhs: Self) {
*self = *self >> rhs;
}
}
}
impl_ref_ops! {
impl core::ops::ShrAssign<$scalar> for $vector {
#[inline]
fn shr_assign(&mut self, rhs: $scalar) {
*self = *self >> rhs;
}
}
}
)*
)*
};
}
/// Implements unsigned integer operators for the provided types.
macro_rules! impl_signed_int_ops {
{ $($scalar:ty => $($vector:ty),*;)* } => {
impl_unsigned_int_ops! { $($scalar => $($vector),*;)* }
$( // scalar
$( // vector
impl_op! { impl Neg for $vector, $scalar }
)*
)*
};
}
impl_unsigned_int_ops! {
u8 => crate::u8x8, crate::u8x16, crate::u8x32, crate::u8x64;
u16 => crate::u16x4, crate::u16x8, crate::u16x16, crate::u16x32;
u32 => crate::u32x2, crate::u32x4, crate::u32x8, crate::u32x16;
u64 => crate::u64x2, crate::u64x4, crate::u64x8;
u128 => crate::u128x2, crate::u128x4;
usize => crate::usizex2, crate::usizex4, crate::usizex8;
}
impl_signed_int_ops! {
i8 => crate::i8x8, crate::i8x16, crate::i8x32, crate::i8x64;
i16 => crate::i16x4, crate::i16x8, crate::i16x16, crate::i16x32;
i32 => crate::i32x2, crate::i32x4, crate::i32x8, crate::i32x16;
i64 => crate::i64x2, crate::i64x4, crate::i64x8;
i128 => crate::i128x2, crate::i128x4;
isize => crate::isizex2, crate::isizex4, crate::isizex8;
}
impl_float_ops! {
f32 => crate::f32x2, crate::f32x4, crate::f32x8, crate::f32x16;
f64 => crate::f64x2, crate::f64x4, crate::f64x8;
}
impl_mask_ops! {
crate::mask8 => crate::mask8x8, crate::mask8x16, crate::mask8x32, crate::mask8x64;
crate::mask16 => crate::mask16x4, crate::mask16x8, crate::mask16x16, crate::mask16x32;
crate::mask32 => crate::mask32x2, crate::mask32x4, crate::mask32x8, crate::mask32x16;
crate::mask64 => crate::mask64x2, crate::mask64x4, crate::mask64x8;
crate::mask128 => crate::mask128x2, crate::mask128x4;
crate::masksize => crate::masksizex2, crate::masksizex4, crate::masksizex8;
}