Stabilize armv8 neon instruction set on aarch64 #1266

SparrowLii · 2021-12-10T02:11:30Z

According to the description in #90972 and the fcp has entered the final comment period stage, I think we can stabilize the armv8 neon instruction set on aarch64.

rust-highfive · 2021-12-10T02:11:34Z

r? @Amanieu

(rust-highfive has picked a reviewer for you, use r? to override)

Amanieu · 2021-12-10T21:53:21Z

crates/core_arch/src/aarch64/crc.rs

@@ -13,6 +13,7 @@ use stdarch_test::assert_instr;
 #[inline]
 #[target_feature(enable = "crc")]
 #[cfg_attr(test, assert_instr(crc32x))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub unsafe fn __crc32d(crc: u32, data: u64) -> u32 {


The FCP only covers the NEON (SIMD) intrinsics, not crc intrinsics.

OK. Unmarked it.

Amanieu · 2021-12-10T21:54:26Z

crates/core_arch/src/aarch64/v8.rs

@@ -11,20 +11,23 @@ use stdarch_test::assert_instr;
 /// Reverse the order of the bytes.
 #[inline]
 #[cfg_attr(test, assert_instr(rev))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub unsafe fn _rev_u64(x: u64) -> u64 {


Same here, there intrinsics are not covered by the FCP.

OK. Unmarked it.

Amanieu · 2021-12-10T21:55:01Z

crates/core_arch/src/aarch64/tme.rs

@@ -71,7 +71,7 @@ pub const _TMFAILURE_TRIVIAL: u64 = 1 << 24;
 /// [ARM TME Intrinsics](https://developer.arm.com/docs/101028/0010/transactional-memory-extension-tme-intrinsics).
 #[inline]
 #[target_feature(enable = "tme")]
-#[cfg_attr(test, assert_instr(tstart))]
+#[cfg_attr(test, assert_instr(nop))]


Why were these changed to nop? This is incorrect.

The assert_instr of these instructions failed on my local aarch64. But since these instructions are not included in this stabilization, I think we can keep them unchanged.

It might have failed because the binutils version on your system is too old and objdump is not able to disassemble the instruction. We use a recent version of ubuntu in the docker image used in CI for this reason.

Amanieu · 2021-12-10T21:56:17Z

crates/stdarch-gen/neon.spec

@@ -7044,7 +7044,7 @@ a = 1.1, 1.9, -1.7, -2.3
 validate 1.0, 2.0, -2.0, -2.0
 target = frintts

-aarch64 = frint32x
+aarch64 = nop


Why is this changed to nop? This is incorrect.

Same as above. Changes have been cancelled.

Amanieu · 2021-12-10T21:56:54Z

crates/core_arch/src/arm_shared/neon/mod.rs

@@ -4811,7 +6687,7 @@ pub unsafe fn vpadalq_u32(a: uint64x2_t, b: uint32x4_t) -> uint64x2_t {
 #[target_feature(enable = "neon")]
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v8"))]
 #[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
-#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smmla))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]


Why is this changed to nop? This is incorrect.

Also why is this function not marked as stable?

Same as above. And vmmla instructions require v8.6a architecture, so they are not included in this stabilization.

Amanieu · 2021-12-10T21:57:16Z

crates/stdarch-gen/src/main.rs

- FPArmV8 => "fp-armv8,v8",
+ ArmV7 => "neon",
+ Vfp4 => "neon",
+ FPArmV8 => "neon",


Why did you change this? The previous target features were correct.

There are two functions used to generate instructions: gen_arm and gen_aarch64. The part I changed is in gen_aarch64. The values of these targets do not affect the target_feature on aarch64, so unified to neon I think it is more appropriate

That's fine if it's only on aarch64.

Amanieu · 2021-12-10T21:58:59Z

We still need to wait ~1 week for the FCP to complete. Once it is complete, this PR can be merged.

Amanieu · 2021-12-11T01:50:47Z

crates/core_arch/src/arm_shared/neon/mod.rs

+#[cfg_attr(
+ target_arch = "aarch64",
+ stable(feature = "neon_intrinsics", since = "1.59.0")
+)]
 pub unsafe fn vext_u64<const N: i32>(a: uint64x1_t, _b: uint64x1_t) -> uint64x1_t {
 if N != 0 {
 unreachable_unchecked()


I just noticed this: we should be using a static assert here, not unreachable_unchecked. This applies to all vext_{s,u,p,f}64.

update stdarch 2 commits in d219ad6..0716b22 2021-12-9 23:50:37 +0000 to 2021-12-14 16:17:57 +0100 * Fix a bunch of typos ([Fix a bunch of typos stdarch#1267](rust-lang/stdarch#1267)) * Stabilize armv8 neon instruction set on aarch64 ([Stabilize armv8 neon instruction set on aarch64 stdarch#1266](rust-lang/stdarch#1266)) The update stabilizes armv8 neon instructions on aarch64. rust-lang#90972

Pkgsrc changes: * Bump available bootstraps to 1.58.1. * Adjust one patch (and checksum) so that it still applies. Upstream changes: Version 1.59.0 (2022-02-24) ========================== Language -------- - [Stabilize default arguments for const generics][90207] - [Stabilize destructuring assignment][90521] - [Relax private in public lint on generic bounds and where clauses of trait impls][90586] - [Stabilize asm! and global_asm! for x86, x86_64, ARM, Aarch64, and RISC-V][91728] Compiler -------- - [Stabilize new symbol mangling format, leaving it opt-in (-Csymbol-mangling-version=v0)][90128] - [Emit LLVM optimization remarks when enabled with `-Cremark`][90833] - [Fix sparc64 ABI for aggregates with floating point members][91003] - [Warn when a `#[test]`-like built-in attribute macro is present multiple times.][91172] - [Add support for riscv64gc-unknown-freebsd][91284] - [Stabilize `-Z emit-future-incompat` as `--json future-incompat`][91535] - [Soft disable incremental compilation][94124] This release disables incremental compilation, unless the user has explicitly opted in via the newly added RUSTC_FORCE_INCREMENTAL=1 environment variable. This is due to a known and relatively frequently occurring bug in incremental compilation, which causes builds to issue internal compiler errors. This particular bug is already fixed on nightly, but that fix has not yet rolled out to stable and is deemed too risky for a direct stable backport. As always, we encourage users to test with nightly and report bugs so that we can track failures and fix issues earlier. See [94124] for more details. [94124]: rust-lang/rust#94124 Libraries --------- - [Remove unnecessary bounds for some Hash{Map,Set} methods][91593] Stabilized APIs --------------- - [`std::thread::available_parallelism`][available_parallelism] - [`Result::copied`][result-copied] - [`Result::cloned`][result-cloned] - [`arch::asm!`][asm] - [`arch::global_asm!`][global_asm] - [`ops::ControlFlow::is_break`][is_break] - [`ops::ControlFlow::is_continue`][is_continue] - [`TryFrom<char> for u8`][try_from_char_u8] - [`char::TryFromCharError`][try_from_char_err] implementing `Clone`, `Debug`, `Display`, `PartialEq`, `Copy`, `Eq`, `Error` - [`iter::zip`][zip] - [`NonZeroU8::is_power_of_two`][is_power_of_two8] - [`NonZeroU16::is_power_of_two`][is_power_of_two16] - [`NonZeroU32::is_power_of_two`][is_power_of_two32] - [`NonZeroU64::is_power_of_two`][is_power_of_two64] - [`NonZeroU128::is_power_of_two`][is_power_of_two128] - [`DoubleEndedIterator for ToLowercase`][lowercase] - [`DoubleEndedIterator for ToUppercase`][uppercase] - [`TryFrom<&mut [T]> for [T; N]`][tryfrom_ref_arr] - [`UnwindSafe for Once`][unwindsafe_once] - [`RefUnwindSafe for Once`][refunwindsafe_once] - [armv8 neon intrinsics for aarch64][stdarch/1266] Const-stable: - [`mem::MaybeUninit::as_ptr`][muninit_ptr] - [`mem::MaybeUninit::assume_init`][muninit_init] - [`mem::MaybeUninit::assume_init_ref`][muninit_init_ref] - [`ffi::CStr::from_bytes_with_nul_unchecked`][cstr_from_bytes] Cargo ----- - [Stabilize the `strip` profile option][cargo/10088] - [Stabilize future-incompat-report][cargo/10165] - [Support abbreviating `--release` as `-r`][cargo/10133] - [Support `term.quiet` configuration][cargo/10152] - [Remove `--host` from cargo {publish,search,login}][cargo/10145] Compatibility Notes ------------------- - [Refactor weak symbols in std::sys::unix][90846] This may add new, versioned, symbols when building with a newer glibc, as the standard library uses weak linkage rather than dynamically attempting to load certain symbols at runtime. - [Deprecate crate_type and crate_name nested inside `#![cfg_attr]`][83744] This adds a future compatibility lint to supporting the use of cfg_attr wrapping either crate_type or crate_name specification within Rust files; it is recommended that users migrate to setting the equivalent command line flags. - [Remove effect of `#[no_link]` attribute on name resolution][92034] This may expose new names, leading to conflicts with preexisting names in a given namespace and a compilation failure. - [Cargo will document libraries before binaries.][cargo/10172] - [Respect doc=false in dependencies, not just the root crate][cargo/10201] - [Weaken guarantee around advancing underlying iterators in zip][83791] - [Make split_inclusive() on an empty slice yield an empty output][89825] - [Update std::env::temp_dir to use GetTempPath2 on Windows when available.][89999] - [unreachable! was updated to match other formatting macro behavior on Rust 2021][92137] Internal Changes ---------------- These changes provide no direct user facing benefits, but represent significant improvements to the internals and overall performance of rustc and related tools. - [Fix many cases of normalization-related ICEs][91255] - [Replace dominators algorithm with simple Lengauer-Tarjan][85013] - [Store liveness in interval sets for region inference][90637] - [Remove `in_band_lifetimes` from the compiler and standard library, in preparation for removing this unstable feature.][91867] [91867]: rust-lang/rust#91867 [83744]: rust-lang/rust#83744 [83791]: rust-lang/rust#83791 [85013]: rust-lang/rust#85013 [89825]: rust-lang/rust#89825 [89999]: rust-lang/rust#89999 [90128]: rust-lang/rust#90128 [90207]: rust-lang/rust#90207 [90521]: rust-lang/rust#90521 [90586]: rust-lang/rust#90586 [90637]: rust-lang/rust#90637 [90833]: rust-lang/rust#90833 [90846]: rust-lang/rust#90846 [91003]: rust-lang/rust#91003 [91172]: rust-lang/rust#91172 [91255]: rust-lang/rust#91255 [91284]: rust-lang/rust#91284 [91535]: rust-lang/rust#91535 [91593]: rust-lang/rust#91593 [91728]: rust-lang/rust#91728 [91878]: rust-lang/rust#91878 [91896]: rust-lang/rust#91896 [91926]: rust-lang/rust#91926 [91984]: rust-lang/rust#91984 [92020]: rust-lang/rust#92020 [92034]: rust-lang/rust#92034 [92483]: rust-lang/rust#92483 [cargo/10088]: rust-lang/cargo#10088 [cargo/10133]: rust-lang/cargo#10133 [cargo/10145]: rust-lang/cargo#10145 [cargo/10152]: rust-lang/cargo#10152 [cargo/10165]: rust-lang/cargo#10165 [cargo/10172]: rust-lang/cargo#10172 [cargo/10201]: rust-lang/cargo#10201 [cargo/10269]: rust-lang/cargo#10269 [cstr_from_bytes]: https://doc.rust-lang.org/stable/std/ffi/struct.CStr.html#method.from_bytes_with_nul_unchecked [muninit_ptr]: https://doc.rust-lang.org/stable/std/mem/union.MaybeUninit.html#method.as_ptr [muninit_init]: https://doc.rust-lang.org/stable/std/mem/union.MaybeUninit.html#method.assume_init [muninit_init_ref]: https://doc.rust-lang.org/stable/std/mem/union.MaybeUninit.html#method.assume_init_ref [unwindsafe_once]: https://doc.rust-lang.org/stable/std/sync/struct.Once.html#impl-UnwindSafe [refunwindsafe_once]: https://doc.rust-lang.org/stable/std/sync/struct.Once.html#impl-RefUnwindSafe [tryfrom_ref_arr]: https://doc.rust-lang.org/stable/std/convert/trait.TryFrom.html#impl-TryFrom%3C%26%27_%20mut%20%5BT%5D%3E [lowercase]: https://doc.rust-lang.org/stable/std/char/struct.ToLowercase.html#impl-DoubleEndedIterator [uppercase]: https://doc.rust-lang.org/stable/std/char/struct.ToUppercase.html#impl-DoubleEndedIterator [try_from_char_err]: https://doc.rust-lang.org/stable/std/char/struct.TryFromCharError.html [available_parallelism]: https://doc.rust-lang.org/stable/std/thread/fn.available_parallelism.html [result-copied]: https://doc.rust-lang.org/stable/std/result/enum.Result.html#method.copied [result-cloned]: https://doc.rust-lang.org/stable/std/result/enum.Result.html#method.cloned [asm]: https://doc.rust-lang.org/stable/core/arch/macro.asm.html [global_asm]: https://doc.rust-lang.org/stable/core/arch/macro.global_asm.html [is_break]: https://doc.rust-lang.org/stable/std/ops/enum.ControlFlow.html#method.is_break [is_continue]: https://doc.rust-lang.org/stable/std/ops/enum.ControlFlow.html#method.is_continue [try_from_char_u8]: https://doc.rust-lang.org/stable/std/primitive.char.html#impl-TryFrom%3Cchar%3E [zip]: https://doc.rust-lang.org/stable/std/iter/fn.zip.html [is_power_of_two8]: https://doc.rust-lang.org/stable/core/num/struct.NonZeroU8.html#method.is_power_of_two [is_power_of_two16]: https://doc.rust-lang.org/stable/core/num/struct.NonZeroU16.html#method.is_power_of_two [is_power_of_two32]: https://doc.rust-lang.org/stable/core/num/struct.NonZeroU32.html#method.is_power_of_two [is_power_of_two64]: https://doc.rust-lang.org/stable/core/num/struct.NonZeroU64.html#method.is_power_of_two [is_power_of_two128]: https://doc.rust-lang.org/stable/core/num/struct.NonZeroU128.html#method.is_power_of_two [stdarch/1266]: rust-lang/stdarch#1266

glandium · 2022-08-10T00:18:29Z

Is it an oversight that many of these now stable functions take arguments with types that still aren't stable? (I say many, but I haven't checked, it could just as well be all of them)

thomcc · 2022-08-10T02:56:02Z

It’s not an oversight, it’s a rustdoc bug. Those are stable on aarch64, but not arm32, so it gets mislabeled (theres an open issue for it, I can’t provide it at the moment though)

SparrowLii added 3 commits December 9, 2021 15:13

test a few stabilization and correct some assert_instr

45dcf31

stabilize aarch64/neon/mod.rs and arm_shared/neon/mod.rs

269dfba

stabilize aarch64/neon/generated.rs and arm_shared/neon/generated.rs

c2840ee

rust-highfive assigned Amanieu Dec 10, 2021

correct fmt

34cf17a

SparrowLii force-pushed the stable_aarch64 branch from 13c8bbd to 34cf17a Compare December 10, 2021 02:16

Amanieu reviewed Dec 10, 2021

View reviewed changes

correct scope of stabilization

5b8e8e6

Amanieu reviewed Dec 11, 2021

View reviewed changes

use static_assert in vext

634cd27

Amanieu merged commit a389fee into rust-lang:master Dec 13, 2021

SparrowLii mentioned this pull request Dec 15, 2021

update stdarch rust-lang/rust#91951

Merged

bartelsielski mentioned this pull request Feb 9, 2022

Stabilize existing neon instrinsics #1125

Open

This was referenced Feb 25, 2022

Tracking Issue for NEON intrinsics rust-lang/rust#90972

Closed

AArch64 SIMD intrinsics are now stable rusticstuff/simdutf8#60

Closed

SIMD on AArch64 is now available on stable Rust RazrFalcon/tiny-skia#46

Closed

zonyitoo mentioned this pull request Dec 20, 2022

服务端升级到shadowsocks-rust v1.15.0之后，客户端shadowsocks-android v5.3.1-preview不能使用 shadowsocks/shadowsocks-android#2966

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stabilize armv8 neon instruction set on aarch64 #1266

Stabilize armv8 neon instruction set on aarch64 #1266

SparrowLii commented Dec 10, 2021

rust-highfive commented Dec 10, 2021

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021

Amanieu Dec 11, 2021

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021 •

edited

Loading

Amanieu Dec 10, 2021

SparrowLii Dec 11, 2021

Amanieu Dec 11, 2021

Amanieu commented Dec 10, 2021

Amanieu Dec 11, 2021

glandium commented Aug 10, 2022

thomcc commented Aug 10, 2022

Stabilize armv8 neon instruction set on aarch64 #1266

Stabilize armv8 neon instruction set on aarch64 #1266

Conversation

SparrowLii commented Dec 10, 2021

rust-highfive commented Dec 10, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SparrowLii Dec 11, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Amanieu commented Dec 10, 2021

Choose a reason for hiding this comment

glandium commented Aug 10, 2022

thomcc commented Aug 10, 2022

SparrowLii Dec 11, 2021 •

edited

Loading