implement vec_round and friends
This commit is contained in:
Folkert de Vries
Amanieu d'Antras
parent
b75bc77065
commit
cefc61d22e
1 changed files with 284 additions and 0 deletions
|
|
@ -73,6 +73,10 @@ unsafe extern "unadjusted" {
|
|||
#[link_name = "llvm.umin.v8i16"] fn vmnlh(a: vector_unsigned_short, b: vector_unsigned_short) -> vector_unsigned_short;
|
||||
#[link_name = "llvm.umin.v4i32"] fn vmnlf(a: vector_unsigned_int, b: vector_unsigned_int) -> vector_unsigned_int;
|
||||
#[link_name = "llvm.umin.v2i64"] fn vmnlg(a: vector_unsigned_long_long, b: vector_unsigned_long_long) -> vector_unsigned_long_long;
|
||||
|
||||
#[link_name = "llvm.s390.vfisb"] fn vfisb(a: vector_float, b: i32, c: i32) -> vector_float;
|
||||
#[link_name = "llvm.s390.vfidb"] fn vfidb(a: vector_double, b: i32, c: i32) -> vector_double;
|
||||
|
||||
}
|
||||
|
||||
impl_from! { i8x16, u8x16, i16x8, u16x8, i32x4, u32x4, i64x2, u64x2, f32x4, f64x2 }
|
||||
|
|
@ -629,6 +633,71 @@ mod sealed {
|
|||
}
|
||||
|
||||
impl_vec_trait! { [VectorOrc vec_orc]+ 2c (orc) }
|
||||
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub trait VectorRound: Sized {
|
||||
unsafe fn vec_round_impl<const N: i32, const MODE: i32>(self) -> Self;
|
||||
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_roundc(self) -> Self {
|
||||
self.vec_round_impl::<4, 0>()
|
||||
}
|
||||
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_round(self) -> Self {
|
||||
// NOTE: simd_round resoles ties by rounding away from zero,
|
||||
// while the vec_round function rounds towards zero
|
||||
self.vec_round_impl::<4, 4>()
|
||||
}
|
||||
|
||||
// NOTE: vec_roundz (vec_round_impl::<4, 5>) is the same as vec_trunc
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_trunc(self) -> Self {
|
||||
simd_trunc(self)
|
||||
}
|
||||
|
||||
// NOTE: vec_roundp (vec_round_impl::<4, 6>) is the same as vec_ceil
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_ceil(self) -> Self {
|
||||
simd_ceil(self)
|
||||
}
|
||||
|
||||
// NOTE: vec_roundm (vec_round_impl::<4, 7>) is the same as vec_floor
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_floor(self) -> Self {
|
||||
simd_floor(self)
|
||||
}
|
||||
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_rint(self) -> Self {
|
||||
self.vec_round_impl::<0, 0>()
|
||||
}
|
||||
}
|
||||
|
||||
// FIXME(vector-enhancements-1) apply the right target feature to all methods
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
impl VectorRound for vector_float {
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_round_impl<const N: i32, const MODE: i32>(self) -> Self {
|
||||
vfisb(self, N, MODE)
|
||||
}
|
||||
}
|
||||
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
impl VectorRound for vector_double {
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
unsafe fn vec_round_impl<const N: i32, const MODE: i32>(self) -> Self {
|
||||
vfidb(self, N, MODE)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Vector element-wise addition.
|
||||
|
|
@ -843,6 +912,125 @@ where
|
|||
a.vec_orc(b)
|
||||
}
|
||||
|
||||
/// Vector floor.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_floor<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_floor()
|
||||
}
|
||||
|
||||
/// Vector ceil.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_ceil<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_ceil()
|
||||
}
|
||||
|
||||
/// Returns a vector containing the truncated values of the corresponding elements of the given vector.
|
||||
/// Each element of the result contains the value of the corresponding element of a, truncated to an integral value.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_trunc<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_trunc()
|
||||
}
|
||||
|
||||
/// Vector round, resolves ties by rounding towards zero.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_round<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_round()
|
||||
}
|
||||
|
||||
/// Returns a vector by using the current rounding mode to round every
|
||||
/// floating-point element in the given vector to integer.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_roundc<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_roundc()
|
||||
}
|
||||
|
||||
/// Returns a vector containing the largest representable floating-point integral values less
|
||||
/// than or equal to the values of the corresponding elements of the given vector.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_roundm<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
// the IBM docs note
|
||||
//
|
||||
// > vec_roundm provides the same functionality as vec_floor, except that vec_roundz would not trigger the IEEE-inexact exception.
|
||||
//
|
||||
// but in practice `vec_floor` also does not trigger that exception, so both are equivalent
|
||||
a.vec_floor()
|
||||
}
|
||||
|
||||
/// Returns a vector containing the smallest representable floating-point integral values greater
|
||||
/// than or equal to the values of the corresponding elements of the given vector.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_roundp<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
// the IBM docs note
|
||||
//
|
||||
// > vec_roundp provides the same functionality as vec_ceil, except that vec_roundz would not trigger the IEEE-inexact exception.
|
||||
//
|
||||
// but in practice `vec_ceil` also does not trigger that exception, so both are equivalent
|
||||
a.vec_ceil()
|
||||
}
|
||||
|
||||
/// Returns a vector containing the truncated values of the corresponding elements of the given vector.
|
||||
/// Each element of the result contains the value of the corresponding element of a, truncated to an integral value.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_roundz<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
// the IBM docs note
|
||||
//
|
||||
// > vec_roundz provides the same functionality as vec_trunc, except that vec_roundz would not trigger the IEEE-inexact exception.
|
||||
//
|
||||
// but in practice `vec_trunc` also does not trigger that exception, so both are equivalent
|
||||
a.vec_trunc()
|
||||
}
|
||||
|
||||
/// Returns a vector by using the current rounding mode to round every floating-point element in the given vector to integer.
|
||||
#[inline]
|
||||
#[target_feature(enable = "vector")]
|
||||
#[unstable(feature = "stdarch_s390x", issue = "135681")]
|
||||
pub unsafe fn vec_rint<T>(a: T) -> T
|
||||
where
|
||||
T: sealed::VectorRound,
|
||||
{
|
||||
a.vec_rint()
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
|
|
@ -852,6 +1040,33 @@ mod tests {
|
|||
use crate::core_arch::simd::*;
|
||||
use stdarch_test::simd_test;
|
||||
|
||||
macro_rules! test_vec_1 {
|
||||
{ $name: ident, $fn:ident, f32x4, [$($a:expr),+], ~[$($d:expr),+] } => {
|
||||
#[simd_test(enable = "vector")]
|
||||
unsafe fn $name() {
|
||||
let a: vector_float = transmute(f32x4::new($($a),+));
|
||||
|
||||
let d: vector_float = transmute(f32x4::new($($d),+));
|
||||
let r = transmute(vec_cmple(vec_abs(vec_sub($fn(a), d)), vec_splats(f32::EPSILON)));
|
||||
let e = m32x4::new(true, true, true, true);
|
||||
assert_eq!(e, r);
|
||||
}
|
||||
};
|
||||
{ $name: ident, $fn:ident, $ty: ident, [$($a:expr),+], [$($d:expr),+] } => {
|
||||
test_vec_1! { $name, $fn, $ty -> $ty, [$($a),+], [$($d),+] }
|
||||
};
|
||||
{ $name: ident, $fn:ident, $ty: ident -> $ty_out: ident, [$($a:expr),+], [$($d:expr),+] } => {
|
||||
#[simd_test(enable = "vector")]
|
||||
unsafe fn $name() {
|
||||
let a: s_t_l!($ty) = transmute($ty::new($($a),+));
|
||||
|
||||
let d = $ty_out::new($($d),+);
|
||||
let r : $ty_out = transmute($fn(a));
|
||||
assert_eq!(d, r);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
macro_rules! test_vec_2 {
|
||||
{ $name: ident, $fn:ident, $ty: ident, [$($a:expr),+], [$($b:expr),+], [$($d:expr),+] } => {
|
||||
test_vec_2! { $name, $fn, $ty -> $ty, [$($a),+], [$($b),+], [$($d),+] }
|
||||
|
|
@ -1059,4 +1274,73 @@ mod tests {
|
|||
[0b11001100, 0b11001100, 0b11001100, 0b11001100],
|
||||
[0b00110011, 0b11110011, 0b00001100, 0b00000000],
|
||||
[!0b11111111, !0b00111111, !0b11000000, !0b11001100] }
|
||||
|
||||
test_vec_1! { test_vec_floor_f32, vec_floor, f32x4,
|
||||
[1.1, 1.9, -0.5, -0.9],
|
||||
[1.0, 1.0, -1.0, -1.0]
|
||||
}
|
||||
|
||||
test_vec_1! { test_vec_floor_f64_1, vec_floor, f64x2,
|
||||
[1.1, 1.9],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
test_vec_1! { test_vec_floor_f64_2, vec_floor, f64x2,
|
||||
[-0.5, -0.9],
|
||||
[-1.0, -1.0]
|
||||
}
|
||||
|
||||
test_vec_1! { test_vec_ceil_f32, vec_ceil, f32x4,
|
||||
[0.1, 0.5, 0.6, 0.9],
|
||||
[1.0, 1.0, 1.0, 1.0]
|
||||
}
|
||||
test_vec_1! { test_vec_ceil_f64_1, vec_ceil, f64x2,
|
||||
[0.1, 0.5],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
test_vec_1! { test_vec_ceil_f64_2, vec_ceil, f64x2,
|
||||
[0.6, 0.9],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
|
||||
// FIXME(vector-enhancements-1)
|
||||
// test_vec_1! { test_vec_round_f32, vec_round, f32x4,
|
||||
// [],
|
||||
// []
|
||||
// }
|
||||
test_vec_1! { test_vec_round_f64_1, vec_round, f64x2,
|
||||
[0.1, 0.5],
|
||||
[0.0, 0.0]
|
||||
}
|
||||
test_vec_1! { test_vec_round_f64_2, vec_round, f64x2,
|
||||
[0.6, 0.9],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
|
||||
// FIXME(vector-enhancements-1)
|
||||
// test_vec_1! { test_vec_roundc_f32, vec_roundc, f32x4,
|
||||
// [],
|
||||
// []
|
||||
// }
|
||||
test_vec_1! { test_vec_roundc_f64_1, vec_roundc, f64x2,
|
||||
[0.1, 0.5],
|
||||
[0.0, 0.0]
|
||||
}
|
||||
test_vec_1! { test_vec_roundc_f64_2, vec_roundc, f64x2,
|
||||
[0.6, 0.9],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
|
||||
// FIXME(vector-enhancements-1)
|
||||
// test_vec_1! { test_vec_rint_f32, vec_rint, f32x4,
|
||||
// [],
|
||||
// []
|
||||
// }
|
||||
test_vec_1! { test_vec_rint_f64_1, vec_rint, f64x2,
|
||||
[0.1, 0.5],
|
||||
[0.0, 0.0]
|
||||
}
|
||||
test_vec_1! { test_vec_rint_f64_2, vec_rint, f64x2,
|
||||
[0.6, 0.9],
|
||||
[1.0, 1.0]
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue