Use AtomicCell<u64> on targets with target_has_atomic less than 64 #286

taiki-e · 2019-10-07T09:25:31Z

Closes #141

.github/workflows/ci.yml

taiki-e · 2019-10-07T09:34:34Z

src/sync/atomic.rs

+// `#[cfg(target_has_atomic = "64")]`.
+// Refs: https://github.com/rust-lang/rust/tree/master/src/librustc_target
+cfg_if::cfg_if! {
+    if #[cfg(not(any(target_arch = "arm", target_arch = "mips", target_arch = "powerpc")))] {


Many target_arch = "arm" targets are target_has_atomic = "64", but only some targets seems different.

arm-linux-androideabi

armv5te-unknown-linux-musleabi

cc #48 Should we support these targets?

I would definitely support arm-linux-androideabi, I'm hard pressed to find a deployment target with armv5 that is relevant for async-std.

taiki-e · 2019-10-07T09:37:55Z

CI failure seems due to the fact that I enabled the test.

ghost · 2019-10-17T19:15:12Z

This PR needs another rebase.

taiki-e · 2019-10-18T09:04:55Z

This PR should not have changed the behavior on macOS, but this test failed on macOS on CI... But I couldn't reproduce this locally.

nbdd0121 · 2019-10-18T09:26:55Z

Would it be better to name it as something else rather than AtomicU64? Rust atomics are guaranteed to be layout-wise identical to their corresponding integer type, so naming it as AtomicU64 may incorrectly represent that.

Also: std::sync::Mutex is too heavy to protect just an integer. It'd be better to have a spinlock instead.

taiki-e · 2019-10-20T09:52:42Z

Would it be better to name it as something else rather than AtomicU64? Rust atomics are guaranteed to be layout-wise identical to their corresponding integer type, so naming it as AtomicU64 may incorrectly represent that.

I don't intend to make these implementations public, so I don't think it's a problem. However, if #355 is merged, we may need to consider this.

Also: std::sync::Mutex is too heavy to protect just an integer. It'd be better to have a spinlock instead.

I agree. And another advantage of this is AtomicU64::new can be const function.

nbdd0121 · 2019-10-20T14:59:53Z

Another thought, would it make sense to do similar thing as crossbeam's AtomicCell which uses a global Mutex instead? This can guarantee layout compatibility with u64, as well as a const new fn.

ghost

oops, wanted to click "request changes"

taiki-e · 2019-11-04T18:44:10Z

I like the idea of using AtomicCell<u64> from crossbeam-channel as a cross-platform version of AtomicU64 :)

@taiki-e let's do that?

I think it needs a new release of crossbeam-utils including crossbeam-rs/crossbeam#437. So filed crossbeam-rs/crossbeam#444.

ghost · 2019-11-06T23:47:44Z

@taiki-e New versions of crossbeam have been published now.

taiki-e · 2019-11-07T22:10:21Z

looks like ci failure is #286 (comment).

ghost · 2019-11-07T22:27:15Z

@taiki-e Can you try rebasing/merging with the master branch to see if that helps?

taiki-e · 2019-11-08T04:53:47Z

Updated. It seems to succeed on the existing targets (linux,macos,windows), but fails on the newly added targets.

faulesocke · 2020-01-09T23:02:11Z

Rebasing this is almost trivial (just remove the Cargo.toml changes). It then applies cleanly and makes async-std work on MIPS and other platforms.

…4 targets

…on x86-64 targets" This reverts commit b3a65e1.

taiki-e · 2020-01-29T10:30:47Z

Hmm, still a test mentioned in #286 (comment) is failed on the newly added targets.

faulesocke · 2020-03-23T13:28:13Z

Why was this closed? It still does not compile on MIPS.

flxo · 2020-03-26T15:26:08Z

I found this PR today while trying to cross compile async-std for arm-linux-androideabi (32bit). There's no u64 atomic support on 32bit arm v7 and thus also the here proposed usage of crossbeam::atomic doesn't solve the problem. The proposal to use AtomicUsize from #451 sounds reasonable to me but it got closed with a reference to here. Maybe someone can elaborate on the state? I'm happy to support on this issue (I would have to brew my own solution/hack anyway...)

kolapapa · 2020-04-01T11:46:28Z

I found this PR today while trying to cross compile async-std for arm-linux-androideabi (32bit). There's no u64 atomic support on 32bit arm v7 and thus also the here proposed usage of crossbeam::atomic doesn't solve the problem. The proposal to use AtomicUsize from #451 sounds reasonable to me but it got closed with a reference to here. Maybe someone can elaborate on the state? I'm happy to support on this issue (I would have to brew my own solution/hack anyway...)

I thought this PR was going to pass, so I closed it first……

taiki-e · 2020-04-26T16:26:48Z

Why was this closed?

Closed because inactive

taiki-e commented Oct 7, 2019

View reviewed changes

.github/workflows/ci.yml Outdated Show resolved Hide resolved

taiki-e commented Oct 7, 2019

View reviewed changes

taiki-e force-pushed the atomic branch from 4630ec0 to dcce474 Compare October 7, 2019 09:39

yoshuawuyts added the enhancement New feature or request label Oct 8, 2019

taiki-e force-pushed the atomic branch 2 times, most recently from 2b036b2 to 72167f0 Compare October 16, 2019 12:43

taiki-e mentioned this pull request Oct 16, 2019

Enable tests on CI #357

Merged

taiki-e force-pushed the atomic branch 2 times, most recently from 2352bff to bade3b7 Compare October 17, 2019 14:36

taiki-e force-pushed the atomic branch 2 times, most recently from bbd761b to 83487da Compare October 18, 2019 05:00

taiki-e mentioned this pull request Oct 18, 2019

expose std::sync::atomic #355

Closed

taiki-e force-pushed the atomic branch 2 times, most recently from 83487da to d6c5377 Compare October 18, 2019 08:55

taiki-e force-pushed the atomic branch 3 times, most recently from 45484e9 to cb197e8 Compare October 23, 2019 12:08

nbdd0121 mentioned this pull request Oct 25, 2019

Add utility type WakerSet to the sync module #390

Merged

taiki-e force-pushed the atomic branch from cb197e8 to a3e618b Compare October 26, 2019 17:14

taiki-e requested a review from a user October 26, 2019 17:31

taiki-e mentioned this pull request Oct 26, 2019

Remove usage of actions-rs/clippy-check #392

Merged

skade added the nominated for 1.0 Nominated for 1.0 release label Oct 28, 2019

taiki-e force-pushed the atomic branch 2 times, most recently from 1d7d149 to 4dbd49b Compare October 30, 2019 16:18

ghost suggested changes Nov 4, 2019

View reviewed changes

taiki-e force-pushed the atomic branch from d84d7cc to 876b7be Compare November 7, 2019 05:35

taiki-e changed the title ~~Use our own AtomicU64 on targets with target_has_atomic less than 64~~ Use AtomicCell<u64> on targets with target_has_atomic less than 64 Nov 7, 2019

taiki-e force-pushed the atomic branch 2 times, most recently from e9abb0c to 745cf87 Compare November 7, 2019 05:51

taiki-e force-pushed the atomic branch 2 times, most recently from b48511f to a80d6ae Compare November 8, 2019 04:44

kolapapa mentioned this pull request Nov 8, 2019

change DYNAMIC_THREAD_COUNT to be an AtomicUsize #451

Closed

yoshuawuyts mentioned this pull request Jan 9, 2020

Build error on MIPS 32bit #664

Open

taiki-e force-pushed the atomic branch from ae976e0 to a2f2671 Compare January 10, 2020 04:20

taiki-e force-pushed the atomic branch from a2f2671 to edc90ab Compare January 20, 2020 02:59

taiki-e added 3 commits January 29, 2020 19:11

Use AtomicCell<u64> on targets with target_has_atomic less than 64

e4f7017

Temporarily ignore io_timeout:: io_timeout_timedout test on non x86-6…

17a41fe

…4 targets

Revert "Temporarily ignore io_timeout:: io_timeout_timedout test on n…

f99d8aa

…on x86-64 targets" This reverts commit b3a65e1.

taiki-e force-pushed the atomic branch from edc90ab to f99d8aa Compare January 29, 2020 10:12

taiki-e closed this Mar 21, 2020

dignifiedquire mentioned this pull request Apr 26, 2020

Test and fix 32 bit targets #756

Merged

jaysonsantos mentioned this pull request May 19, 2020

async-std won't compile on armv5te because of AtomicU64 #789

Closed

taiki-e deleted the atomic branch September 9, 2020 07:54

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use AtomicCell<u64> on targets with target_has_atomic less than 64 #286

Use AtomicCell<u64> on targets with target_has_atomic less than 64 #286

taiki-e commented Oct 7, 2019 •

edited

Loading

taiki-e Oct 7, 2019

taiki-e Oct 7, 2019

skade Oct 7, 2019

taiki-e commented Oct 7, 2019 •

edited

Loading

ghost commented Oct 17, 2019

taiki-e commented Oct 18, 2019

nbdd0121 commented Oct 18, 2019

taiki-e commented Oct 20, 2019

nbdd0121 commented Oct 20, 2019

ghost left a comment

taiki-e commented Nov 4, 2019 •

edited

Loading

ghost commented Nov 6, 2019

taiki-e commented Nov 7, 2019

ghost commented Nov 7, 2019

taiki-e commented Nov 8, 2019

faulesocke commented Jan 9, 2020

taiki-e commented Jan 29, 2020 •

edited

Loading

faulesocke commented Mar 23, 2020

flxo commented Mar 26, 2020

kolapapa commented Apr 1, 2020

taiki-e commented Apr 26, 2020

Use AtomicCell<u64> on targets with target_has_atomic less than 64 #286

Use AtomicCell<u64> on targets with target_has_atomic less than 64 #286

Conversation

taiki-e commented Oct 7, 2019 • edited Loading

taiki-e Oct 7, 2019

Choose a reason for hiding this comment

taiki-e Oct 7, 2019

Choose a reason for hiding this comment

skade Oct 7, 2019

Choose a reason for hiding this comment

taiki-e commented Oct 7, 2019 • edited Loading

ghost commented Oct 17, 2019

taiki-e commented Oct 18, 2019

nbdd0121 commented Oct 18, 2019

taiki-e commented Oct 20, 2019

nbdd0121 commented Oct 20, 2019

ghost left a comment

Choose a reason for hiding this comment

taiki-e commented Nov 4, 2019 • edited Loading

ghost commented Nov 6, 2019

taiki-e commented Nov 7, 2019

ghost commented Nov 7, 2019

taiki-e commented Nov 8, 2019

faulesocke commented Jan 9, 2020

taiki-e commented Jan 29, 2020 • edited Loading

faulesocke commented Mar 23, 2020

flxo commented Mar 26, 2020

kolapapa commented Apr 1, 2020

taiki-e commented Apr 26, 2020

taiki-e commented Oct 7, 2019 •

edited

Loading

taiki-e commented Oct 7, 2019 •

edited

Loading

taiki-e commented Nov 4, 2019 •

edited

Loading

taiki-e commented Jan 29, 2020 •

edited

Loading