1768394 - OOL traps regresses some microbenchmarks (was: Regression on AWFY around 21 April mark on "float-mm.c.*" tests)

Status: RESOLVED WONTFIX

(Core :: JavaScript: WebAssembly, defect, P3)

Description

[Mayank Bansal]

Look at AWFY, on the following benchmarks :

float-mm.c-Average
float-mm.c-First
float-mm.c-Geometric
float-mm.c-Worst

Regression range : https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=201252c849f427b49422fffea6ca65e36cdade1d&tochange=c95859201259666efcd49533fe7cca1ddea5ab24

Comment 1

[Jan de Mooij [:jandem]]

https://arewefastyet.com/win10/benchmarks/raptor-desktop-jetstream2?numDays=60

I think these are WebAssembly benchmarks. Bug 1680243 would be most likely then.

Comment 2

[Alexandru Ionescu (needinfo me) [:alexandrui]]

Is it me you wanted to needinfo?

Comment 3

[Lars T Hansen [:lth]]

Could be. gcc-loops, quicksort, and zlib all improved on the same date.

Microbenchmarks are easily perturbed, and of course this change also perturbed branch prediction because it changed code layout (that was the point), so it's not surprising if some programs are not happy about it. I can take a look later to see if there's anything actionable.

Comment 4

[BugBot [:suhaib / :marco/ :calixte]]

The severity field is not set for this bug.
:rhunt, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 5

[Mayank Bansal]

Should we close this bug and move on?

Comment 6

[Ryan Hunt [:rhunt]]

Yeah, we should close this.