Reformat std_detect

This commit is contained in:
Jakub Beránek
2025-07-04 10:08:58 +02:00
parent 1057a2213a
commit ee6f2c5276
12 changed files with 54 additions and 182 deletions

View File

@@ -3,9 +3,7 @@
#![allow(dead_code)] // not used on all platforms
use core::sync::atomic::Ordering;
use core::sync::atomic::AtomicUsize;
use core::sync::atomic::{AtomicUsize, Ordering};
/// Sets the `bit` of `x`.
#[inline]
@@ -40,20 +38,14 @@ impl Initializer {
/// Tests the `bit` of the cache.
#[inline]
pub(crate) fn test(self, bit: u32) -> bool {
debug_assert!(
bit < CACHE_CAPACITY,
"too many features, time to increase the cache size!"
);
debug_assert!(bit < CACHE_CAPACITY, "too many features, time to increase the cache size!");
test_bit(self.0, bit)
}
/// Sets the `bit` of the cache.
#[inline]
pub(crate) fn set(&mut self, bit: u32) {
debug_assert!(
bit < CACHE_CAPACITY,
"too many features, time to increase the cache size!"
);
debug_assert!(bit < CACHE_CAPACITY, "too many features, time to increase the cache size!");
let v = self.0;
self.0 = set_bit(v, bit);
}
@@ -61,10 +53,7 @@ impl Initializer {
/// Unsets the `bit` of the cache.
#[inline]
pub(crate) fn unset(&mut self, bit: u32) {
debug_assert!(
bit < CACHE_CAPACITY,
"too many features, time to increase the cache size!"
);
debug_assert!(bit < CACHE_CAPACITY, "too many features, time to increase the cache size!");
let v = self.0;
self.0 = unset_bit(v, bit);
}
@@ -73,11 +62,7 @@ impl Initializer {
/// This global variable is a cache of the features supported by the CPU.
// Note: the third slot is only used in x86
// Another Slot can be added if needed without any change to `Initializer`
static CACHE: [Cache; 3] = [
Cache::uninitialized(),
Cache::uninitialized(),
Cache::uninitialized(),
];
static CACHE: [Cache; 3] = [Cache::uninitialized(), Cache::uninitialized(), Cache::uninitialized()];
/// Feature cache with capacity for `size_of::<usize>() * 8 - 1` features.
///
@@ -104,19 +89,14 @@ impl Cache {
#[inline]
pub(crate) fn test(&self, bit: u32) -> Option<bool> {
let cached = self.0.load(Ordering::Relaxed);
if cached == 0 {
None
} else {
Some(test_bit(cached as u128, bit))
}
if cached == 0 { None } else { Some(test_bit(cached as u128, bit)) }
}
/// Initializes the cache.
#[inline]
fn initialize(&self, value: usize) -> usize {
debug_assert_eq!((value & !Cache::MASK), 0);
self.0
.store(value | Cache::INITIALIZED_BIT, Ordering::Relaxed);
self.0.store(value | Cache::INITIALIZED_BIT, Ordering::Relaxed);
value
}
}
@@ -217,7 +197,5 @@ pub(crate) fn test(bit: u32) -> bool {
} else {
(bit - 2 * Cache::CAPACITY, 2)
};
CACHE[idx]
.test(relative_bit)
.unwrap_or_else(|| detect_and_initialize().test(bit))
CACHE[idx].test(relative_bit).unwrap_or_else(|| detect_and_initialize().test(bit))
}

View File

@@ -29,7 +29,6 @@ mod arch;
#[doc(hidden)]
#[unstable(feature = "stdarch_internal", issue = "none")]
pub use self::arch::__is_feature_detected;
pub(crate) use self::arch::Feature;
mod bit;

View File

@@ -17,9 +17,10 @@
//! - [Linux documentation](https://www.kernel.org/doc/Documentation/arm64/cpu-feature-registers.txt)
//! - [ARM documentation](https://developer.arm.com/documentation/ddi0601/2022-12/AArch64-Registers?lang=en)
use crate::detect::{Feature, cache};
use core::arch::asm;
use crate::detect::{Feature, cache};
/// Try to read the features from the system registers.
///
/// This will cause SIGILL if the current OS is not trapping the mrs instruction.
@@ -104,10 +105,7 @@ pub(crate) fn parse_system_registers(
let sha2 = bits_shift(aa64isar0, 15, 12) >= 1;
enable_feature(Feature::sha2, asimd && sha1 && sha2);
enable_feature(Feature::rdm, asimd && bits_shift(aa64isar0, 31, 28) >= 1);
enable_feature(
Feature::dotprod,
asimd && bits_shift(aa64isar0, 47, 44) >= 1,
);
enable_feature(Feature::dotprod, asimd && bits_shift(aa64isar0, 47, 44) >= 1);
enable_feature(Feature::sve, asimd && bits_shift(aa64pfr0, 35, 32) >= 1);
}

View File

@@ -2,9 +2,10 @@
//!
//! <https://developer.apple.com/documentation/kernel/1387446-sysctlbyname/determining_instruction_set_characteristics>
use crate::detect::{Feature, cache};
use core::ffi::CStr;
use crate::detect::{Feature, cache};
#[inline]
fn _sysctlbyname(name: &CStr) -> bool {
use libc;
@@ -14,13 +15,7 @@ fn _sysctlbyname(name: &CStr) -> bool {
let enabled_ptr = &mut enabled as *mut i32 as *mut libc::c_void;
let ret = unsafe {
libc::sysctlbyname(
name.as_ptr(),
enabled_ptr,
&mut enabled_len,
core::ptr::null_mut(),
0,
)
libc::sysctlbyname(name.as_ptr(), enabled_ptr, &mut enabled_len, core::ptr::null_mut(), 0)
};
match ret {

View File

@@ -54,11 +54,8 @@ fn archauxv(key: libc::c_int) -> usize {
// https://github.com/freebsd/freebsd-src/blob/release/11.4.0/sys/sys/auxv.h
// FreeBSD 11 support in std has been removed in Rust 1.75 (https://github.com/rust-lang/rust/pull/114521),
// so we can safely use this function.
let res = libc::elf_aux_info(
key,
&mut out as *mut libc::c_ulong as *mut libc::c_void,
OUT_LEN,
);
let res =
libc::elf_aux_info(key, &mut out as *mut libc::c_ulong as *mut libc::c_void, OUT_LEN);
// If elf_aux_info fails, `out` will be left at zero (which is the proper default value).
debug_assert!(res == 0 || out == 0);
}

View File

@@ -343,14 +343,8 @@ impl AtHwcap {
enable_feature(Feature::sve2, sve2);
enable_feature(Feature::sve2p1, self.sve2p1 && sve2);
// SVE2 extensions require SVE2 and crypto features
enable_feature(
Feature::sve2_aes,
self.sveaes && self.svepmull && sve2 && self.aes,
);
enable_feature(
Feature::sve2_sm4,
self.svesm4 && sve2 && self.sm3 && self.sm4,
);
enable_feature(Feature::sve2_aes, self.sveaes && self.svepmull && sve2 && self.aes);
enable_feature(Feature::sve2_sm4, self.svesm4 && sve2 && self.sm3 && self.sm4);
enable_feature(
Feature::sve2_sha3,
self.svesha3 && sve2 && self.sha512 && self.sha3 && self.sha1 && self.sha2,

View File

@@ -1,8 +1,9 @@
//! Run-time feature detection for LoongArch on Linux.
use core::arch::asm;
use super::auxvec;
use crate::detect::{Feature, bit, cache};
use core::arch::asm;
/// Try to read the features from the auxiliary vector.
pub(crate) fn detect_features() -> cache::Initializer {
@@ -43,16 +44,8 @@ pub(crate) fn detect_features() -> cache::Initializer {
//
// [hwcap]: https://github.com/torvalds/linux/blob/master/arch/loongarch/include/uapi/asm/hwcap.h
if let Ok(auxv) = auxvec::auxv() {
enable_feature(
&mut value,
Feature::f,
bit::test(cpucfg2, 1) && bit::test(auxv.hwcap, 3),
);
enable_feature(
&mut value,
Feature::d,
bit::test(cpucfg2, 2) && bit::test(auxv.hwcap, 3),
);
enable_feature(&mut value, Feature::f, bit::test(cpucfg2, 1) && bit::test(auxv.hwcap, 3));
enable_feature(&mut value, Feature::d, bit::test(cpucfg2, 2) && bit::test(auxv.hwcap, 3));
enable_feature(&mut value, Feature::lsx, bit::test(auxv.hwcap, 4));
enable_feature(&mut value, Feature::lasx, bit::test(auxv.hwcap, 5));
enable_feature(

View File

@@ -119,16 +119,7 @@ fn _riscv_hwprobe(out: &mut [riscv_hwprobe]) -> bool {
cpus: *mut libc::c_ulong,
flags: libc::c_uint,
) -> libc::c_long {
unsafe {
libc::syscall(
__NR_riscv_hwprobe,
pairs,
pair_count,
cpu_set_size,
cpus,
flags,
)
}
unsafe { libc::syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_set_size, cpus, flags) }
}
unsafe { __riscv_hwprobe(out.as_mut_ptr(), out.len(), 0, ptr::null_mut(), 0) == 0 }

View File

@@ -4,8 +4,10 @@
//! https://github.com/openbsd/src/commit/d335af936b9d7dd9cf655cae1ce19560c45de6c8
//! https://github.com/golang/go/commit/cd54ef1f61945459486e9eea2f016d99ef1da925
use core::mem::MaybeUninit;
use core::ptr;
use crate::detect::cache;
use core::{mem::MaybeUninit, ptr};
// Defined in machine/cpu.h.
// https://github.com/openbsd/src/blob/72ccc03bd11da614f31f7ff76e3f6fce99bc1c79/sys/arch/arm64/include/cpu.h#L25-L40

View File

@@ -4,7 +4,6 @@
use core::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::*;
use core::mem;
use crate::detect::{Feature, bit, cache};
@@ -42,12 +41,7 @@ pub(crate) fn detect_features() -> cache::Initializer {
// 0x8000_0000]. - The vendor ID is stored in 12 u8 ascii chars,
// returned in EBX, EDX, and ECX (in that order):
let (max_basic_leaf, vendor_id) = unsafe {
let CpuidResult {
eax: max_basic_leaf,
ebx,
ecx,
edx,
} = __cpuid(0);
let CpuidResult { eax: max_basic_leaf, ebx, ecx, edx } = __cpuid(0);
let vendor_id: [[u8; 4]; 3] = [ebx.to_ne_bytes(), edx.to_ne_bytes(), ecx.to_ne_bytes()];
let vendor_id: [u8; 12] = mem::transmute(vendor_id);
(max_basic_leaf, vendor_id)
@@ -60,11 +54,8 @@ pub(crate) fn detect_features() -> cache::Initializer {
// EAX = 1, ECX = 0: Queries "Processor Info and Feature Bits";
// Contains information about most x86 features.
let CpuidResult {
ecx: proc_info_ecx,
edx: proc_info_edx,
..
} = unsafe { __cpuid(0x0000_0001_u32) };
let CpuidResult { ecx: proc_info_ecx, edx: proc_info_edx, .. } =
unsafe { __cpuid(0x0000_0001_u32) };
// EAX = 7: Queries "Extended Features";
// Contains information about bmi,bmi2, and avx2 support.
@@ -76,11 +67,8 @@ pub(crate) fn detect_features() -> cache::Initializer {
extended_features_edx_leaf_1,
) = if max_basic_leaf >= 7 {
let CpuidResult { ebx, ecx, edx, .. } = unsafe { __cpuid(0x0000_0007_u32) };
let CpuidResult {
eax: eax_1,
edx: edx_1,
..
} = unsafe { __cpuid_count(0x0000_0007_u32, 0x0000_0001_u32) };
let CpuidResult { eax: eax_1, edx: edx_1, .. } =
unsafe { __cpuid_count(0x0000_0007_u32, 0x0000_0001_u32) };
(ebx, ecx, edx, eax_1, edx_1)
} else {
(0, 0, 0, 0, 0) // CPUID does not support "Extended Features"
@@ -89,10 +77,7 @@ pub(crate) fn detect_features() -> cache::Initializer {
// EAX = 0x8000_0000, ECX = 0: Get Highest Extended Function Supported
// - EAX returns the max leaf value for extended information, that is,
// `cpuid` calls in range [0x8000_0000; u32::MAX]:
let CpuidResult {
eax: extended_max_basic_leaf,
..
} = unsafe { __cpuid(0x8000_0000_u32) };
let CpuidResult { eax: extended_max_basic_leaf, .. } = unsafe { __cpuid(0x8000_0000_u32) };
// EAX = 0x8000_0001, ECX=0: Queries "Extended Processor Info and Feature
// Bits"
@@ -208,10 +193,8 @@ pub(crate) fn detect_features() -> cache::Initializer {
// Processor Extended State Enumeration Sub-leaf (EAX = 0DH,
// ECX = 1):
if max_basic_leaf >= 0xd {
let CpuidResult {
eax: proc_extended_state1_eax,
..
} = unsafe { __cpuid_count(0xd_u32, 1) };
let CpuidResult { eax: proc_extended_state1_eax, .. } =
unsafe { __cpuid_count(0xd_u32, 1) };
enable(proc_extended_state1_eax, 0, Feature::xsaveopt);
enable(proc_extended_state1_eax, 1, Feature::xsavec);
enable(proc_extended_state1_eax, 3, Feature::xsaves);
@@ -269,10 +252,8 @@ pub(crate) fn detect_features() -> cache::Initializer {
enable(extended_features_edx_leaf_1, 8, Feature::amx_complex);
if max_basic_leaf >= 0x1e {
let CpuidResult {
eax: amx_feature_flags_eax,
..
} = unsafe { __cpuid_count(0x1e_u32, 1) };
let CpuidResult { eax: amx_feature_flags_eax, .. } =
unsafe { __cpuid_count(0x1e_u32, 1) };
enable(amx_feature_flags_eax, 4, Feature::amx_fp8);
enable(amx_feature_flags_eax, 5, Feature::amx_transpose);