user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

s390x vector: add vec_cntlz, vec_cnttz and vec_popcnt

Browse source
This commit is contained in:
Folkert de Vries 2025-01-22 21:09:26 +01:00 • committed by Amanieu d'Antras
parent 46fbfe9b09
commit 507a298ba9
2 changed files with 114 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

8
library/stdarch/crates/core_arch/src/s390x/macros.rs
View file

@ -2,6 +2,14 @@
#![allow(unused_imports)] // FIXME remove when more tests are added
macro_rules! test_impl {
($fun:ident +($($v:ident : $ty:ty),*) -> $r:ty [$call:ident, $instr:ident]) => {
#[inline]
#[target_feature(enable = "vector")]
#[cfg_attr(test, assert_instr($instr))]
pub unsafe fn $fun ($($v : $ty),*) -> $r {
transmute($call ($($v),*))
}
};
($fun:ident ($($v:ident : $ty:ty),*) -> $r:ty [$call:ident, $instr:ident]) => {
#[inline]
#[target_feature(enable = "vector")]

106
library/stdarch/crates/core_arch/src/s390x/vector.rs
View file

@ -450,8 +450,79 @@ mod sealed {
(u64, u64x2, vector_bool_long_long),
(i64, i64x2, vector_bool_long_long)
}
#[unstable(feature = "stdarch_s390x", issue = "135681")]
pub trait CountBits {
type Result;
unsafe fn vec_cntlz(self) -> Self::Result;
unsafe fn vec_cnttz(self) -> Self::Result;
unsafe fn vec_popcnt(self) -> Self::Result;
}
macro_rules! impl_count_bits {
($ty:tt) => {
#[unstable(feature = "stdarch_s390x", issue = "135681")]
impl CountBits for $ty {
type Result = t_u!($ty);
#[inline]
#[target_feature(enable = "vector")]
unsafe fn vec_cntlz(self) -> Self::Result {
transmute(simd_ctlz(self))
}
#[inline]
#[target_feature(enable = "vector")]
unsafe fn vec_cnttz(self) -> Self::Result {
transmute(simd_cttz(self))
}
#[inline]
#[target_feature(enable = "vector")]
unsafe fn vec_popcnt(self) -> Self::Result {
transmute(simd_ctpop(self))
}
}
};
}
impl_count_bits!(vector_signed_char);
impl_count_bits!(vector_unsigned_char);
impl_count_bits!(vector_signed_short);
impl_count_bits!(vector_unsigned_short);
impl_count_bits!(vector_signed_int);
impl_count_bits!(vector_unsigned_int);
impl_count_bits!(vector_signed_long_long);
impl_count_bits!(vector_unsigned_long_long);
test_impl! { vec_clzb_signed +(a: vector_signed_char) -> vector_unsigned_char [simd_ctlz, vclzb] }
test_impl! { vec_clzh_signed +(a: vector_signed_short) -> vector_unsigned_short [simd_ctlz, vclzh] }
test_impl! { vec_clzf_signed +(a: vector_signed_int) -> vector_unsigned_int [simd_ctlz, vclzf] }
test_impl! { vec_clzg_signed +(a: vector_signed_long_long) -> vector_unsigned_long_long [simd_ctlz, vclzg] }
test_impl! { vec_clzb_unsigned +(a: vector_unsigned_char) -> vector_unsigned_char [simd_ctlz, vclzb] }
test_impl! { vec_clzh_unsigned +(a: vector_unsigned_short) -> vector_unsigned_short [simd_ctlz, vclzh] }
test_impl! { vec_clzf_unsigned +(a: vector_unsigned_int) -> vector_unsigned_int [simd_ctlz, vclzf] }
test_impl! { vec_clzg_unsigned +(a: vector_unsigned_long_long) -> vector_unsigned_long_long [simd_ctlz, vclzg] }
test_impl! { vec_ctzb_signed +(a: vector_signed_char) -> vector_unsigned_char [simd_cttz, vctzb] }
test_impl! { vec_ctzh_signed +(a: vector_signed_short) -> vector_unsigned_short [simd_cttz, vctzh] }
test_impl! { vec_ctzf_signed +(a: vector_signed_int) -> vector_unsigned_int [simd_cttz, vctzf] }
test_impl! { vec_ctzg_signed +(a: vector_signed_long_long) -> vector_unsigned_long_long [simd_cttz, vctzg] }
test_impl! { vec_ctzb_unsigned +(a: vector_unsigned_char) -> vector_unsigned_char [simd_cttz, vctzb] }
test_impl! { vec_ctzh_unsigned +(a: vector_unsigned_short) -> vector_unsigned_short [simd_cttz, vctzh] }
test_impl! { vec_ctzf_unsigned +(a: vector_unsigned_int) -> vector_unsigned_int [simd_cttz, vctzf] }
test_impl! { vec_ctzg_unsigned +(a: vector_unsigned_long_long) -> vector_unsigned_long_long [simd_cttz, vctzg] }
// FIXME(vector-enhancements-1) other integer types are emulated, but get their own
// instructions in later facilities. Add tests when possible.
test_impl! { vec_popcnt_signed +(a: vector_signed_char) -> vector_signed_char [simd_ctpop, vpopctb] }
test_impl! { vec_popcnt_unsigned +(a: vector_unsigned_char) -> vector_unsigned_char [simd_ctpop, vpopctb] }
}
/// Vector element-wise addition.
#[inline]
#[target_feature(enable = "vector")]
@ -491,6 +562,41 @@ where
a.vec_mul(b)
}
/// Vector Count Leading Zeros
#[inline]
#[target_feature(enable = "vector")]
#[unstable(feature = "stdarch_s390x", issue = "135681")]
pub unsafe fn vec_cntlz<T>(a: T) -> <T as sealed::CountBits>::Result
where
T: sealed::CountBits,
{
a.vec_cntlz()
}
/// Vector Count Trailing Zeros
#[inline]
#[target_feature(enable = "vector")]
#[unstable(feature = "stdarch_s390x", issue = "135681")]
pub unsafe fn vec_cnttz<T>(a: T) -> <T as sealed::CountBits>::Result
where
T: sealed::CountBits,
{
a.vec_cnttz()
}
/// Vector Population Count
///
/// Computes the population count (number of set bits) in each element of the input.
#[inline]
#[target_feature(enable = "vector")]
#[unstable(feature = "stdarch_s390x", issue = "135681")]
pub unsafe fn vec_popcnt<T>(a: T) -> <T as sealed::CountBits>::Result
where
T: sealed::CountBits,
{
a.vec_popcnt()
}
/// Vector element-wise maximum.
#[inline]
#[target_feature(enable = "vector")]