Implement prefetch hints for aarch64

[dgbo]

Co-authored-by: Wang Maozhang [email protected]

This implements prefetch hints for aarch64 via LLVMs prefetch instrinsic.

We wintness ~20% performance gain in seal_pre_commit_phase1 of Filecoin with 32GB data input.

Verified with cargo +nightly test and cargo +nightly test --release.
There are four failures (as shown below), but these are not cause by this PR.
These tests failed both with and without this PR.

---- core_arch::aarch64::tme::assert___tcommit_tcommit stdout ----
disassembly for stdarch_test_shim___tcommit_tcommit:
         0: adrp x8, 16f000 <_GLOBAL_OFFSET_TABLE_+0x280>
         1: ldr x8, [x8, #3792]
         2: adrp x9, f1000 <anon.1626de38755f4729e9f2a08512645c0a.14.llvm.15569108828795629504+0x10b87>
         3: add x9, x9, #0xd29
         4: str x9, [x8]
         5: msr s0_3_c3_c0_3, xzr
         6: ret
thread 'core_arch::aarch64::tme::assert___tcommit_tcommit' panicked at 'failed to find instruction `tcommit` in the disassembly', crates/stdarch-test/src/lib.rs:144:9


failures:
    core_arch::aarch64::tme::assert___tcancel_tcancel
    core_arch::aarch64::tme::assert___tcommit_tcommit
    core_arch::aarch64::tme::assert___tstart_tstart
    core_arch::aarch64::tme::assert___ttest_ttest

Thanks.

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @Amanieu (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

[dgbo]

The check error is strange:

error[E0723]: can only call `transmute` from const items, not `const fn`
   --> crates/core_arch/src/wasm32/simd128.rs:420:5
    |
420 |     transmute(f64x2(a0, a1))
    |     ^^^^^^^^^^^^^^^^^^^^^^^^
    |
    = note: see issue #57563 <https://github.com/rust-lang/rust/issues/57563> for more information
    = help: add `#![feature(const_fn)]` to the crate attributes to enable

This PR does not touch any code in wasm32 at all.
And the #![feature(const_fn)] is in crates/core_arch/src/lib.rs already.

Could anyone give some suggestion here?

Best Regards.

[Amanieu]

Use #[rustc_args_required_const(1, 2)] to ensure that rw and locality are compile-time constants.

[bjorn3]

You also need to match on every valid (rw, locality) pair so that the prefetch call always gets constant values as arguments and you need to handle invalid values.

[dgbo]

Thanks a lot for the quick review.

The #[rustc_args_required_const(1, 2)] is added just before the function _prefetch.

Created a macro pref to match all valid (rw, locality) pairs.
An invalid pair will cause panic now.
Suggestions?

[Amanieu]

The wasm failure seems to be due to an issue on the latest Rust nightly. You can fix it by adding the const_fn_transmute feature to the list in lib.rs.

[dgbo]

The wasm failure seems to be due to an issue on the latest Rust nightly. You can fix it by adding the const_fn_transmute feature to the list in lib.rs.

Added the const_fn_transmute feature to the list in lib.rs. The CI is green. Thanks.

[Amanieu]

Thanks!