1603886 - Assertion failure: in->canProduceFloat32() || (in->resultTypeSet() && in->resultTypeSet()->empty()), at js/src/jit/IonAnalysis.cpp:1833

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

The following testcase crashes on mozilla-central revision f09f24f2b545 (build with --enable-debug --disable-optimize, run with --fuzzing-safe --ion-eager):

function g(x, y) {
    y ? 1 : 1;
}
function h(x) {
    g(1 ? 1 : d(t | undefined), Math.round() ? x : Math.fround(Math.hypot(undefined | Math.round)));
}
let z = [[], false];
for (var j = 0; j < 4; ++j) {
    for (var k = 0; k < 99; ++k) {
        h(z[j]);
    }
}

Backtrace:

#0  0x0000561910bbc896 in (anonymous namespace)::TypeAnalyzer::adjustPhiInputs (this=0x7f0267301da0, phi=0x7f0267be1268) at js/src/jit/IonAnalysis.cpp:1832
#1  0x0000561910bba5ce in (anonymous namespace)::TypeAnalyzer::insertConversions (this=0x7f0267301da0) at js/src/jit/IonAnalysis.cpp:1967
#2  0x0000561910b460cd in (anonymous namespace)::TypeAnalyzer::analyze (this=0x7f0267301da0) at js/src/jit/IonAnalysis.cpp:2255
#3  0x0000561910b3e489 in js::jit::ApplyTypeInformation (mir=0x7f0267bdb300, graph=...) at js/src/jit/IonAnalysis.cpp:2267
#4  0x0000561910b377e5 in js::jit::OptimizeMIR (mir=0x7f0267bdb300) at js/src/jit/Ion.cpp:1198
#5  0x0000561910b40821 in js::jit::CompileBackEnd (mir=0x7f0267bdb300) at js/src/jit/Ion.cpp:1630
#6  0x0000561910b5909b in js::jit::IonBuilder::runTask (this=0x7f0267bdb2f8) at js/src/jit/IonBuilder.cpp:1181
#7  0x000056190fce6535 in js::HelperThread::handleIonWorkload (this=0x7f0267b08d20, locked=...) at js/src/vm/HelperThreads.cpp:2278
/snip

For detailed crash information, see attachment.

Setting s-s as a start as MIR is on the stack. Note that this is fairly intermittent.

Comment 1

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Attachment: Detailed Crash Information

Comment 2

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Pernosco link:

https://pernos.co/debug/4jzAfEJa3VoGEIeK75qCGQ/index.html

Oh, and the testcase is a little intermittent but fairly reproducible.

Comment 3

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

autobisectjs shows this is probably related to the following changeset:

The first bad revision is:
changeset: https://hg.mozilla.org/mozilla-central/rev/7e8fdfa9aa32
user: Jan de Mooij
date: Wed Dec 11 07:58:30 2019 +0000
summary: Bug 1602190 - Fix bug in float32 specialization of phis. r=bbouvier

Jan, bug 1602190 probably added the assertion that caused this issue, so which bug would be the real regressor?

Comment 4

[Jan de Mooij [:jandem]]

The assertion I added isn't sound in some cases, it can fail after conversion instructions have been inserted. Not sure how to fix it - I think I'll have to remove the assertion for now.

Comment 5

[Jan de Mooij [:jandem]]

Attachment: Bug 1603886 - Remove an assertion that doesn't always hold. r?bbouvier!

I added this assertion, but it can fail after conversion/box instructions have
been added.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/0bfce120565d
Remove an assertion that doesn't always hold. r=bbouvier

Comment 7

[Andrei Ciure[:aciure]]

https://hg.mozilla.org/mozilla-central/rev/0bfce120565d