Bug 291410

Summary: MotionMark regression (287234@main) 0.76%, Canvas Lines 6.5%
Product: WebKit Reporter: Gerald Squelart <g_squelart>
Component: CSSAssignee: Sam Weinig <sam>
Status: ASSIGNED    
Severity: Normal CC: g_squelart, koivisto
Priority: P2 Keywords: InRadar
Version: WebKit Nightly Build   
Hardware: Unspecified   
OS: Unspecified   
Bug Depends on: 283828    
Bug Blocks:    

Gerald Squelart
Reported 2025-04-10 23:49:42 PDT
Running MotionMark on iMac20,1 (Intel, 27in, 2020) shows this regression at 287234@main https://bugs.webkit.org/show_bug.cgi?id=283828 : (overall) A: 2.548 Kpt ± 3.1 pt B: 2.529 Kpt ± 5.1 pt 0.76% worse (significant with 98% probability) Canvas Arcs A: 7.514 Kpt ± 16 pt B: 7.523 Kpt ± 8.6 pt 0.12% better (insignificant) Canvas Lines A: 45.52 Kpt ± 0.30 Kpt B: 42.55 Kpt ± 0.19 Kpt 6.54% worse (significant with 98% probability) Design A: 472.6 pt ± 3.6 pt B: 476.4 pt ± 8.4 pt 0.81% better (insignificant) Images A: 423.9 pt ± 1.1 pt B: 423.2 pt ± 1.7 pt 0.16% worse (insignificant) Leaves A: 3.093 Kpt ± 13 pt B: 3.070 Kpt ± 13 pt 0.73% worse (significant with 98% probability) Multiply A: 2.924 Kpt ± 9.6 pt B: 2.955 Kpt ± 8.4 pt 1.06% better (significant with 98% probability) Paths A: 2.727 Kpt ± 11 pt B: 2.714 Kpt ± 12 pt 0.45% worse (insignificant) Suits A: 1.053 Kpt ± 3.7 pt B: 1.053 Kpt ± 3.1 pt 0.06% better (insignificant) The regression was smaller on more modern hardware, maybe a quarter of that on the 2025 M4 MacBook Air. It's quite a complex patch (to me), so Sam could you please have a look, see if you could guess or track down what could have caused this regression?
Attachments
Gerald Squelart
Comment 1 2025-04-10 23:50:01 PDT
Note You need to log in before you can comment on or make changes to this bug.