3.2_f32 formatted as 0 on powerpc64le #96306

NobodyXu · 2022-04-22T04:23:21Z

In my PR ParkMyCar/compact_str#16 , when formatting 3.2 using std, it gives me 0 on target powerpc64le instead of 3.2 (it works on all other target).

P.S. ryu also format it as 0.0 on it dtolnay/ryu#47

mbartlett21 · 2022-04-23T10:34:52Z

Did you mean dtolnay/ryu#47 for the ryu issue?

@rustbot label +O-PowerPC

NobodyXu · 2022-04-23T10:36:38Z

Did you mean dtolnay/ryu#47 for the ryu issue?

Oh yes, thank you for pointing this out

NobodyXu · 2022-04-23T15:17:36Z

I just tried to reproduce this on my own x86_64 linux box by running:

cross +nightly test --release --all-features --manifest-path=compact_str/Cargo.toml --target powerpc-unknown-linux-gnu

however, all tests passed without problem...

But the test running on github action still failed on powerpc64le

nagisa · 2022-04-23T22:35:44Z

Marked this as I-unsound though I don’t really have a definite culprit here. Historically these sorts of problems have been caused by calling convention mismatches, codegen failures or some other similarly scary and “has potential to cause other code to break much more terribly” problem.

nagisa · 2022-04-23T22:36:58Z

I just tried to reproduce this on my own x86_64 linux box by running:

You used a 32-bit PPC target here. Does this reproduce with a powerpc64le target?

cross +nightly test --release --all-features --manifest-path=compact_str/Cargo.toml --target powerpc64le-unknown-linux-gnu

NobodyXu · 2022-04-24T02:13:40Z

cross +nightly test --release --all-features --manifest-path=compact_str/Cargo.toml --target powerpc64le-unknown-linux-gnu

Thanks for pointing this out again!

With the updated command, I was able to reproduce that.

NobodyXu · 2022-04-24T02:15:37Z

Now I was also able to reproduce it using this:

fn main() {
    println!("{}", 3.2_f32);
}

And it gives me 0 using cross +nightly run --release --target powerpc64le-unknown-linux-gnu (the --release flag can be omitted).

NobodyXu · 2022-04-24T09:44:56Z

This is the assembly I get:

0000000000008d50 <tests::main>:
    8d50: 09 00 4c 3c  	addis 2, 12, 9
    8d54: b0 f1 42 38  	addi 2, 2, -3664
    8d58: a6 02 08 7c  	mflr 0
    8d5c: 10 00 01 f8  	std 0, 16(1)
    8d60: 51 ff 21 f8  	stdu 1, -176(1)
    8d64: fd ff 62 3c  	addis 3, 2, -3
    8d68: 6c c4 63 38  	addi 3, 3, -15252
    8d6c: 4d fb ff 4b  	bl 0x88b8
    8d70: 00 00 00 60  	nop
    8d74: 60 00 81 f8  	std 4, 96(1)
    8d78: 68 00 61 f8  	std 3, 104(1)
    8d7c: 60 00 61 e8  	ld 3, 96(1)
    8d80: 68 00 81 e8  	ld 4, 104(1)
    8d84: a0 00 81 f8  	std 4, 160(1)
    8d88: a8 00 61 f8  	std 3, 168(1)
    8d8c: a0 00 c1 38  	addi 6, 1, 160
    8d90: ff ff 62 3c  	addis 3, 2, -1
    8d94: c0 48 83 38  	addi 4, 3, 18624
    8d98: 70 00 61 38  	addi 3, 1, 112
    8d9c: 02 00 a0 38  	li 5, 2
    8da0: 01 00 e0 38  	li 7, 1
    8da4: c5 fd ff 4b  	bl 0x8b68
    8da8: 00 00 00 60  	nop
    8dac: 70 00 61 38  	addi 3, 1, 112
    8db0: 59 77 01 48  	bl 0x20508
    8db4: 00 00 00 60  	nop
    8db8: b0 00 21 38  	addi 1, 1, 176
    8dbc: 10 00 01 e8  	ld 0, 16(1)
    8dc0: a6 03 08 7c  	mtlr 0
    8dc4: 20 00 80 4e  	blr
		...
    8dd4: 00 00 00 60  	nop
    8dd8: 00 00 00 60  	nop
    8ddc: 00 00 00 60  	nop

apiraino · 2022-05-03T15:18:46Z

Assigning priority as discussed in the Zulip thread of the Prioritization Working Group.

@rustbot label -I-prioritize +P-critical

pnkfelix · 2022-05-04T13:17:37Z

@NobodyXu any chance you can run your test on older versions of the compiler? If this is a regression, then one can even use a tool like https://github.com/rust-lang/cargo-bisect-rustc to identify the point where the regression was injected.

NobodyXu · 2022-05-04T14:10:42Z

@NobodyXu any chance you can run your test on older versions of the compiler? If this is a regression, then one can even use a tool like https://github.com/rust-lang/cargo-bisect-rustc to identify the point where the regression was injected.

Yes, but since I don't have a power pc, ~~I will have to use cross and cargo-hack to perform regression testing~~ as cross does not work wit cargo-bisect-rustc cross-rs/cross#699

NobodyXu · 2022-05-04T14:24:16Z

It seems that the bug is fixed on cargo 1.62.0-nightly (f63f23ff1 2022-04-28), ran using command cross +nightly run --target powerpc64le-unknown-linux-gnu (using --release does not change this behavior).

NobodyXu · 2022-05-04T14:28:03Z

cargo 1.60.0 (d1fd9fe 2022-03-01) also passed the test, ran using command cross run --target powerpc64le-unknown-linux-gnu (using --release does not change this behavior).

NobodyXu · 2022-05-05T05:35:48Z

I think we need more regression tests in rustc, having float parsing being broken and randomly fixed isn't a sustainable approach to support powerpc64le.

NobodyXu · 2022-05-05T06:15:05Z

There are still some strange bugs for float on powerpc64le here:

    assert_eq!("-inf", f32::NEG_INFINITY.to_compact_str());

failed with:

thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `"-inf"`,
 right: `"inf"`', compact_str/src/tests.rs:664:5

This works on all other platforms except for powerpc64le.

apiraino · 2022-05-05T14:43:59Z

Re-evaluating priority as per T-compiler meeting on Zulip

@rustbot label +P-high -P-critical

pnkfelix · 2022-05-05T14:44:06Z

Those are definitely some worrisome points you raise, @NobodyXu !

I'm hoping someone can help investigate them, and it would certainly be great to flesh out our tests here some more

NobodyXu · 2022-05-15T09:49:53Z

There are still some strange bugs for float on powerpc64le here:
    assert_eq!("-inf", f32::NEG_INFINITY.to_compact_str());
failed with:
thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `"-inf"`,
 right: `"inf"`', compact_str/src/tests.rs:664:5
This works on all other platforms except for powerpc64le.

On cargo 1.62.0-nightly (3f052d8ee 2022-05-12), the problem described above is fixed.

Edit:

According to the CI log, it's reintroduced on 1.63.0-nightly (42e1761c7 2022-05-15).

NobodyXu · 2022-05-16T05:24:01Z

@pnkfelix IMO we absolutely need more unit tests for it, as I just witnessed the regression of f32::NEG_INFINITY again on 1.63.0-nightly.

This strange regression that coming back makes it impossible to use compact_str or other crates that depends on the float formatting on powerpc64le reliably.

Edit:

Fixed again on cargo 1.63.0-nightly (3f052d8ee 2022-05-12)

Emilgardis · 2022-06-16T03:14:29Z

The issues mentioned here might've been issues with QEMU 3.0.1 and fixed with QEMU 5.1.0 (or earlier, 5.1.0 is the version used by latest unreleased cross today), see discussion in cross-rs/cross#803 (comment)

It seems this specific case has not been tested on real hardware, but on newer QEMU versions this works just fine.

EDIT:
For anyone that is hitting this issue while using cross, I highly recommend you try out the new version.

if you're using the Github action actions-rs/cargo you can try out the newer version of cross by adding

- run: cargo install --git https://github.com/cross-rs/cross

before your cargo test/run/build invocation.

We're hoping to land these changes this month so that this will not have to be done.

NobodyXu · 2022-06-16T03:17:41Z

Out of curiosity, does rust-lang run CI on actual powerpc machine?

If not, then is this the reason for powerpc to be a tier 2 platform?

nagisa · 2022-06-16T15:17:21Z

We don't run CI on a PowerPC host at all, emulated or baremetal, afaik. There are insufficient engineering resources for setting everything up and maintaining and there hasn't been a motivated party who would pick this sort of work up either. 16 Jun 2022 06:17:53 Jiahao XU ***@***.***>:

…

Out of curiosity, does rust-lang run CI on actual powerpc machine? If not, then is this the reason for powerpc to be a tier 2 platform? -- Reply to this email directly or view it on GitHub: #96306 (comment) You are receiving this because you commented. Message ID: ***@***.***>

nagisa · 2022-06-16T16:05:36Z

FWIW I’ve just tested the affected crate on a real ppc64le-linux machine running a POWER9 chip. All of 1.58.0, 1.59.0, 1.60.0, 1.61.0, nightly-2022-05-{11,15} successfully ran

cargo +$VER test --release --all-features --manifest-path=compact_str/Cargo.toml --target powerpc64le-unknown-linux-gnu -- --include-ignored

to completion at both master and 0ed53d042202550091a71980458d18cbb3919e64.

NobodyXu · 2022-06-17T03:16:33Z

FWIW I’ve just tested the affected crate on a real ppc64le-linux machine running a POWER9 chip. All of 1.58.0, 1.59.0, 1.60.0, 1.61.0, nightly-2022-05-{11,15} successfully ran
cargo +$VER test --release --all-features --manifest-path=compact_str/Cargo.toml --target powerpc64le-unknown-linux-gnu -- --include-ignored
to completion at both master and 0ed53d042202550091a71980458d18cbb3919e64.

It currently has tests disabled on powerpc, you would need to manually enable it.

NobodyXu · 2022-06-17T03:21:10Z

@nagisa You would need to remove the cfg here (not ignoreed).

The implementation uses crate ryu to provide better performance.
It will still be able to detect compiler bugs on powerpc though, just that sometimes it gives even stranger bugs.

NobodyXu · 2022-06-17T03:30:10Z

@nagisa Checkout this repository I used to reproduce the bug.

nagisa · 2022-06-17T10:11:00Z

As I wrote in the comment above, I tested the commit 0ed53d042202550091a71980458d18cbb3919e64 which removes the cfg in question as well. Was that insufficient?

NobodyXu · 2022-06-17T11:39:42Z

Was that insufficient?

That's sufficient.
Sorry I missed that part.

pnkfelix · 2022-11-11T15:03:52Z

Discussed in P-high review

It appears that this problem is resolved, according to the comments here.

Therefore, closing.

NobodyXu added the C-bug Category: This is a bug. label Apr 22, 2022

rustbot added the O-PowerPC Target: PowerPC processors label Apr 23, 2022

nagisa added the I-unsound Issue: A soundness hole (worst kind of bug), see: https://en.wikipedia.org/wiki/Soundness label Apr 23, 2022

rustbot added the I-prioritize Issue: Indicates that prioritization has been requested for this issue. label Apr 23, 2022

nagisa added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Apr 23, 2022

This was referenced Apr 25, 2022

Add && Impl new trait ToCompactStr ParkMyCar/compact_str#16

Merged

tests: Ignore failing test on PowerPC 64-bit ParkMyCar/compact_str#83

Merged

rustbot added P-critical Critical priority and removed I-prioritize Issue: Indicates that prioritization has been requested for this issue. labels May 3, 2022

rustbot added P-high High priority and removed P-critical Critical priority labels May 5, 2022

pnkfelix closed this as completed Nov 11, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

3.2_f32 formatted as 0 on powerpc64le #96306

3.2_f32 formatted as 0 on powerpc64le #96306

NobodyXu commented Apr 22, 2022 •

edited

mbartlett21 commented Apr 23, 2022

NobodyXu commented Apr 23, 2022 •

edited

NobodyXu commented Apr 23, 2022

nagisa commented Apr 23, 2022

nagisa commented Apr 23, 2022

NobodyXu commented Apr 24, 2022

NobodyXu commented Apr 24, 2022 •

edited

NobodyXu commented Apr 24, 2022

apiraino commented May 3, 2022

pnkfelix commented May 4, 2022

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 5, 2022

NobodyXu commented May 5, 2022 •

edited

apiraino commented May 5, 2022 •

edited

pnkfelix commented May 5, 2022

NobodyXu commented May 15, 2022 •

edited

NobodyXu commented May 16, 2022 •

edited

Emilgardis commented Jun 16, 2022 •

edited

NobodyXu commented Jun 16, 2022

nagisa commented Jun 16, 2022 via email

nagisa commented Jun 16, 2022

NobodyXu commented Jun 17, 2022

NobodyXu commented Jun 17, 2022 •

edited

NobodyXu commented Jun 17, 2022

nagisa commented Jun 17, 2022

NobodyXu commented Jun 17, 2022

pnkfelix commented Nov 11, 2022

3.2_f32 formatted as 0 on powerpc64le #96306

3.2_f32 formatted as 0 on powerpc64le #96306

Comments

NobodyXu commented Apr 22, 2022 • edited

mbartlett21 commented Apr 23, 2022

NobodyXu commented Apr 23, 2022 • edited

NobodyXu commented Apr 23, 2022

nagisa commented Apr 23, 2022

nagisa commented Apr 23, 2022

NobodyXu commented Apr 24, 2022

NobodyXu commented Apr 24, 2022 • edited

NobodyXu commented Apr 24, 2022

apiraino commented May 3, 2022

pnkfelix commented May 4, 2022

NobodyXu commented May 4, 2022 • edited

NobodyXu commented May 4, 2022 • edited

NobodyXu commented May 4, 2022 • edited

NobodyXu commented May 5, 2022

NobodyXu commented May 5, 2022 • edited

apiraino commented May 5, 2022 • edited

pnkfelix commented May 5, 2022

NobodyXu commented May 15, 2022 • edited

NobodyXu commented May 16, 2022 • edited

Emilgardis commented Jun 16, 2022 • edited

NobodyXu commented Jun 16, 2022

nagisa commented Jun 16, 2022 via email

nagisa commented Jun 16, 2022

NobodyXu commented Jun 17, 2022

NobodyXu commented Jun 17, 2022 • edited

NobodyXu commented Jun 17, 2022

nagisa commented Jun 17, 2022

NobodyXu commented Jun 17, 2022

pnkfelix commented Nov 11, 2022

NobodyXu commented Apr 22, 2022 •

edited

NobodyXu commented Apr 23, 2022 •

edited

NobodyXu commented Apr 24, 2022 •

edited

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 4, 2022 •

edited

NobodyXu commented May 5, 2022 •

edited

apiraino commented May 5, 2022 •

edited

NobodyXu commented May 15, 2022 •

edited

NobodyXu commented May 16, 2022 •

edited

Emilgardis commented Jun 16, 2022 •

edited

NobodyXu commented Jun 17, 2022 •

edited